Upstart Intro, Cookbook and Best Practises


1   Meta

1.1   Document Version

This is document edit 218.

See footer for further details.

1.2   Authors


1.3   Acknowledgements

The Authors are grateful to the following individuals who have provided valuable input to this document:

  • Colin Watson (Canonical)
  • Scott James Remnant (Canonical, Google), author of Upstart.
  • James Page (Canonical)
  • Joel Ebel (Google)
  • Mark Russell (Canonical)
  • Bradley Ayers
  • Kenneth Porter
  • Roberto Alsina (Canonical), reStructuredText Guru.

1.4   Purpose

The purpose of this document is multi-faceted. It is intended as:

  • A gentle introduction to Upstart.
  • A Cookbook of recipes and best-practises for solving common and not so common problems.
  • An extended guide to the configuration syntax of Upstart.

It attempts to explain the intricacies of Upstart with worked examples and lots of details.

Note that the reference documentation for Upstart will always be the manual pages: this is merely a supplement to them.

1.5   Suggestions and Errata

Bad documentation is often worse than no documentation. If you find a problem with this document, however small...

  • spelling error
  • grammatical error
  • factual error
  • inconsistency
  • lack of clarity
  • ambiguous or misleading content
  • missing information
  • et cetera

... or if you'd like to see some particular feature covered please raise a bug report on the Upstart Cookbook project website so that we can improve this work:

As an incentive you will be credited in the Acknowledgements section.

1.6   Coverage

There are essentially two major versions of Upstart covered by this document:

1.6.1   Upstream Upstart

This is the pure, or "vanilla" version which is designed to work on any Linux system:

1.6.2   Debian and Ubuntu Version of Upstart

The Debian and Ubuntu-packaged version [17].

This is a "debianised" version of Upstart (in other words, a version packaged for Debian and derivatives). It includes a few minor changes specifically for running Upstart on Debian and Ubuntu systems, namely:

  • Change to the way the console is initialised, to work with Plymouth.
  • Initramfs to root filesystem context hand-off changes.


1.6.3   Availability

Upstart is relied upon by millions of systems across a number of different Operating Systems including:

  • Google's Chrome OS
  • Google's Chromium OS
  • Red Hat's RHEL 6 [38]
  • Ubuntu

It is also available as an option for other systems such as:

1.6.4   Releases

Table of official Upstart releases:

Upstart Releases.
Release Date Version
2009-07-21 0.6.2
2009-08-02 0.6.3
2010-02-04 0.6.5
2010-04-27 0.6.6
2010-12-14 0.6.7
2011-03-01 1.0
2011-03-16 1.1
2011-03-22 1.2
2011-06-14 1.3
2011-12-13 1.4
2012-03-22 1.5
2012-11-15 1.6
2012-12-07 1.6.1
2013-03-04 1.7
2013-03-22 1.8
2013-06-28 1.9
2013-07-04 1.9.1
2013-08-23 1.10
2013-11-14 1.11
2014-03-07 1.12
2014-03-11 1.12.1
2014-07-11 1.13
2014-07-16 1.13.1
2014-09-04 1.13.2

1.6.5   Debian-specific and Ubuntu-Specific Content ( , )

This document is written with Debian and Ubuntu in mind, but will attempt to identify Debian-specific and Ubuntu-specific behaviour where appropriate by showing one of these icons (or both if the content applies to both Debian and Ubuntu):

  • Debian-specific icon: (displays as "D" on section headings).
  • Ubuntu-specific icon: (displays as "U" on section headings).

1.7   Audience

This document is targeted at:

  • Users interested in learning about Upstart.
  • System Administrators looking to make the most of the capabilities of Upstart.
  • Developers and Packagers who wish to package their application to work with Upstart.

1.8   Document Preparation

This document is written in reStructuredText, a textual markup language. The document was prepared using the following tools:

1.10   Warning

This document aims to aid understanding of Upstart and identify some hopefully useful "canned" solutions and advice to common problems and questions.

The authors have taken as much care as possible in the preparation of this document. However, you are advised strongly to exercise extreme caution when changing critical system facilities such as the init daemon. Most situations are recoverable and advice is provided in this document, but if your system explodes in a ball of fire or becomes unusable as a result of a suggestion from this document, you alone have the intellectual pleasure of fixing your systems.

2   Typographical Conventions

2.1   Commands and configuration stanzas

Throughout this document a fixed-width font such as this will be used to denote commands, brief command output and configuration stanzas.

2.2   User Input and Command Output

An indented block will be used to denote user input and command output.

2.2.1   Non-Privileged User

Indented lines starting with a dollar character ('$') are used to denote the shell prompt (followed by optional commands) for a non-privileged user. Command output is shown by indented lines not preceded by the dollar character:

$ echo hello

2.2.2   Super-User

Indented lines starting with a hash (or "pound") character ('#') are used to denote the shell prompt (followed by optional commands) for the root user. Command output is shown by indented lines not preceded by the hash character [14]:

# whoami

Note that some examples make use of sudo(8) to show the command should be run as root: the example above could thus be written:

$ sudo whoami

This latter approach is clearer in the context where a comment is also specified using the hash character.

2.3   Configuration Examples

An indented block is also used to show examples of job configuration:

  # a config file
end script

3   Introduction

3.1   What is Upstart?

Quoting from,

Upstart is an event-based replacement for the /sbin/init daemon which handles starting of tasks and services during boot, stopping them during shutdown and supervising them while the system is running.

The "init" or "system initialisation" process on Unix and Linux systems has process ID (PID) "1". That is to say, it is the first process to start when the system boots (ignoring the initrd/initramfs). As the quote shows, Upstart is an "init" replacement for the traditional Unix "System V" "init" system. Upstart provides the same facilities as the traditional "init" system, but surpasses it in many ways.

3.1.1   Reliability

Upstart is written using the NIH Utility Library ("libnih"). This is a very small, efficient and safe library of generic routines. It is designed for applications that run early in the boot sequence ("plumbing"). Reliability and safety is critically important for an init daemon since:

  • it runs as the super-user.
  • it is responsible for managing critical system services.
  • if init exits for any reason, the kernel panics.

To help ensure reliability and avoid regressions, Upstart and the NIH Utility Library both come with comprehensive test suites. See Unit Tests for further information.

3.1.2   Design History

Upstart was created due to fundamental limitations in existing systems. Those systems can be categorized into two types:

  • System V init system
  • Dependency-based init systems

To understand why Upstart was written and why its revolutionary design was chosen, it is necessary to consider these two classes of init system.   Critique of the System V init System   SysV Benefits   Simplicity

Creating service files is easy with SystemV init since they are simply shell scripts. To enable/disable a service in a particular runlevel, you only need to create/remove a symbolic link in a particular directory or set of directories.   Guaranteed Ordering of Services

This is achieved by init running the scripts pointed to by the symbolic links in sequence. The relative order in which init invokes these scripts is determined by a numeric element in the name: lower numbered services run before higher numbered services.   SysV Limitations   Non-Optimal Performance

The traditional sequential boot system was appropriate for the time it was invented, but by modern standards it is "slow" in the sense that it makes no use of parallelism.

It was designed to be simple and efficient for Administrators to manage. However, this model does not make full use of modern system resources, particularly once it is recognised that multiple services can often be run simultaneously.

A common "hack" used by Administrators is to circumvent the serialisation by running their service in the background, such that some degree of parallelism is possible. The fact that this hack is required and is common on such systems demonstrates clearly the flaw in that system.   Server-Centric

In the days of colossal Unix systems with hundreds of concurrent users, where reboots were rare, the traditional SysV approach was perfect. If hardware needed replacing, a system shutdown was scheduled, the shutdown performed, the new hardware was installed and the system was brought back on-line.

However, the world has now moved on. From an Ubuntu perspective, a significant proportion of users run the desktop edition on portable devices where they may reboot multiple times a day.   Assumes Static Hardware at all Times

Modern Linux systems can deal with new hardware devices being added and removed dynamically ("hot-plug"). The traditional SysV init system itself is incapable of handling such a dynamically changing system.   Every Service Does Heavy Lifting

Most service files are fairly formulaic. For example, they might:

  • perform initial checks, such as:
    • ensuring no other instance of a daemon is running.
    • checking the existence of a directory or file.
    • removing old cache files.
  • ensure dependent daemons are running.
  • spawn the main service.

The most difficult and time costly operation these services perform is that of handling dependent daemons. The LSB specifies helper utilities that these services can make use of, but arguably each service shouldn't need to be handling this activity themselves: the init system itself should do it on behalf of the services it manages.   Critique of Dependency-Based init Systems   Benefits of Dependency-based init   Recognises Services Require Other Services

The recognition that services often need to make use of other services is an important improvement over SystemV init systems. It places a bigger responsibility on the init system itself and reduces the complexity and work that needs to be performed by individual service files.   Limitations of Dependency-based init   Does Not Recognise Dynamic Nature of Linux

The main problem with dependency-based init systems is that they approach the problem from the "wrong direction". Again, this is due to their not recognising the dynamic nature of modern Linux systems.

For example, if a dependency-based init system wished to start say MySQL, it would first start all the dependent services that MySQL needed. This sounds perfectly reasonable.

However, consider how such a system would approach the problem of dealing with a user who plugs in an external monitor. Maybe we'd like our system to display some sort of configuration dialogue so the user can choose how they want to use their new monitor in combination with their existing laptop display. This can only be "hacked" with a dependency-based init system since you do not know when the new screen will be plugged. So, your choices are either:

  • Do nothing.

    Corresponds to an inability to handle this scenario.

  • Have a daemon that hangs around polling for new hardware being plugged.

    Wasteful and inefficient.

What you really want is a system that detects such asynchronous events and when the conditions are right for a service to run, the service is started.

This can be summarised as:

  • Upstart starts a service when its required conditions are met.

    The service (job configuration file) only needs to specify the conditions that allow the service to run, and the executable to run the service itself.

  • Dependency-based init systems meet a service's dependencies before starting them.

    Each service generally does this using a brute-force approach of forcing all the dependencies to start.

    Note that the init system itself is not doing the heavy-lifting: that is left up to each service itself (!)

This summary is worth considering carefully as the distinction between the two types of system is subtle but important.

The other problem with dependency-based init systems is that they require a dependency-solver which is often complex and not always optimal.   Upstart's Design: Why It Is Revolutionary

It was necessary to outline the limitations of the SysV and dependency-based init systems to appreciate why Upstart is special...

Upstart is revolutionary as it recognises and was designed specifically for a dynamic system. It handles asynchronicity by emitting events. This too is revolutionary.

Upstart emits "events" which services can register an interest in. When an event -- or combination of events -- is emitted that satisfies some service's requirements, Upstart will automatically start or stop that service. If multiple jobs have the same "start on" condition, Upstart will start those jobs ''in parallel''. To be manifest: Upstart handles starting the "dependent" services itself - this is not handled by the service file itself as it is with dependency-based systems.

Further, Upstart is being guided by the ultimate arbiter of hardware devices: the kernel.

In essence, Upstart is an event engine: it creates events, handles the consequences of those events being emitted and starts and stops processes as required. Like the best Unix software, it does this job very well. It is efficient, fast, flexible and reliable. It makes use of "helper" daemons (such as the upstart-udev-bridge and the upstart-socket-bridge) to inject new types of events into the system and react to these events. This design is sensible and clean: the init system itself must not be compromised since if it fails, the kernel panics. Therefore, any functionality which is not considered "core" functionality is farmed out to other daemons.

See [35] for further details.

3.1.3   Performance

Upstart was designed with performance in mind. It makes heavy use of the NIH Utility Library which is optimised for efficient early boot environments. Additionally, Upstart's design is lightweight, efficient and elegant. At its heart it is an event-based messaging system that has the ability to control and monitor processes. Upstart is designed to manage services running in parallel. It will only start services when the conditions they have specified are met.

3.1.4   Server

Upstart is used by Ubuntu for the Ubuntu Desktop and for Ubuntu Server (and as a result of this, it is also used in the Ubuntu Cloud). Why is Upstart also compelling in a server environment?   Boot Performance

Some say that boot performance is not important on servers, possibly since the time taken to bring RAID arrays on-line is significantly longer than the time it takes to boot the operating system. However, nobody seriously wants their system to take longer than necessary to boot.

Consider also the case for Cloud deployments, which of course run on servers. Here, boot speed is very important as it affects the time taken to deploy a new server instance. The faster you can deploy new services to handle an increasing workload the better the experience for your customers.   Failure Modes

It's a fact that systems and software are getting more complex. In the old days of Unix, runlevels encompassed every major mode of operation you might want your system to handle. However, expectations have changed. Nowadays, we expect systems to react to problems (and maybe even "self-heal" the simple ones).

The landscape has changed and Upstart is fully able to accommodate such changes since its design is clean, elegant and abstract. Crucially, Upstart is not tied to the rigid runlevel system. Indeed, Upstart has no knowledge of runlevels internally, but it supports them trivially with events. And since events are so abstract, they are highly flexible building blocks for higher-level constructs. Added to which, since Upstart's events are dynamic, the system can be configured for a myriad of possible system behaviours and failure modes and have it react accordingly.

4   Concepts and Terminology

The main concepts in Upstart are "events" and "jobs". Understanding the difference between the two is crucial.

4.1   Job

A "unit of work" - generally either a "Task" or a "Service". Jobs are defined in a Job configuration file.

4.1.1   Job Types   Task Job

A Task Job is one which runs a short-running process, that is, a program which might still take a long time to run, but which has a definite lifetime and end state.

For example, deleting a file could be a Task Job since the command starts, deletes the file in question (which might take some time if the file is huge) and then the delete command ends.

In this book Task Jobs are often referred to as tasks.   Service Job

A Service Job is a long-running (or daemon(3) process). It is the opposite of a Task Job since a Service Job might never end of its own accord.

Examples of Service Jobs are entities such as databases, webservers or ftp servers.   Abstract Job

There is one other type of job which has no script sections or exec stanzas. Such abstract jobs can still be started and stopped, but will have no corresponding child process (PID). In fact, starting such a job will result in it "running" perpetually if not stopped by an Administrator. Abstract jobs exist only within Upstart itself but can be very useful. See for example:

4.1.2   Job States

The table below shows all possible Job States and the legal transitions between them. States are exposed to users via the status field in the output of the initctl status command.

Job State Transitions.



start stop
waiting starting n/a
starting security stopping
security pre-start stopping
pre-start spawned stopping
spawned post-start stopping
post-start running stopping
running stopping pre-stop or stopping [15]
pre-stop running stopping
stopping killed killed
killed post-stop post-stop
post-stop starting waiting

For example, if the job is currently in state starting, and its goal is start, it will then move to the pre-start state.

Note that jobs may change state so quickly that you may not be able to observe all the values above in the initctl output. However, you will see the transitions if you raise the log-priority to debug or info. See initctl log-priority for details.

Details of states:

  • waiting : initial state.
  • starting : job is about to start.
  • security : job is having its AppArmor security policy loaded (see apparmor load).
  • pre-start : running pre-start section.
  • spawned : about to run script or exec section.
  • post-start : running post-start section.
  • running : interim state set after post-start section processed denoting job is running (But it may have no associated PID!)
  • pre-stop : running pre-stop section.
  • stopping : interim state set after pre-stop section processed.
  • killed : job is about to be stopped.
  • post-stop : running post-stop section.

State transitions diagram for versions of Upstart up to and including version 1.12.1 (green lines represent goal=start, red lines represent goal=stop):

upstart state diagram

State transitions (up to and including Upstart version 1.12.1).

State transitions diagram for Upstart version 1.13 and newer (green lines represent goal=start, red lines represent goal=stop):

upstart state diagram

State transitions (Upstart version 1.13 and upwards).   Viewing State Transitions

To view state transitions:

  1. Change the log-priority to debug
  2. "tail -f" your system log file
  3. start/stop/restart a job or emit an event.

4.1.3   Job Environment

When Upstart runs a job, it provides it with a very restrictive environment which contains just two system variables:

  • TERM
  • PATH

Upstart itself will also potentially set some special variables the job can use. See Standard Environment Variables for further details.

If your system job needs further variables to be set, you can use the env and export stanzas.

Session Jobs are different. They too can use env and export, but they already inherit the environment of the Session Init that is supervising them. However, further to that, Session Jobs can also influence the environment of the processes that comprise both a single job and all subsequent jobs. See the "env" commands in the initctl Commands Summary for details.

4.2   Job Configuration File

A Job is defined in a Job Configuration File (or more simply a conf file) which is a plain text file containing one or more stanzas. Job configuration files are named:


Where "<name>" should reflect the application being run or the service being provided.

Job configuration files can exist in two types of location, depending on whether they are a System Job or a User Job.

Note that it is common to refer to a Job configuration file as a "job", although technically a job is a running instance of a Job configuration file.

4.2.1   System Job

All system jobs by default live in the following directory:


This directory can be overridden by specifying the --confdir=<directory> option to the init daemon, however this is a specialist option which users should not need to use.

4.2.2   User Job

Deprecated as of Upstart v1.7: see Session Job.

Upstart 1.3 introduced user jobs, allowing non-privileged users to create jobs by placing job configuration files in the following directory:


This feature is not currently enabled in Ubuntu (up to and including 11.10 ("Oneiric Ocelot")).

The syntax for such jobs is identical for "system jobs".


Currently, a user job cannot be created with the same name as a system job: the system job will take precedence.

Controlling user jobs is the same as for system jobs: use initctl, start, stop, et cetera.


Stanzas which manipulate resources limits (such as limit, nice, and oom) may cause a job to fail to start should the value provided to such a stanza attempt to exceed the maximum value the users privilege level allows.


User jobs cannot currently take advantage of job logging. If a user job does specify console log, it is considered to have specified console none. Logging of user jobs is planned for the next release of Upstart.   Enabling

To enable user jobs, the administrator must modify the D-Bus configuration file "Upstart.conf" to allow non-root users access to all the Upstart D-Bus methods and properties. On an Ubuntu system the file to modify is:


The Upstream Upstart 1.3 distribution already includes a "Upstart.conf" file containing the required changes.

4.2.3   Session Job

As of Upstart v1.7

Session Jobs are analogous to the old User Jobs. Unlike the old User Jobs, Session Jobs are not managed by Upstart running as PID 1 - they are managed by the users own Session Init.

Unlike when Upstart runs as PID 1, a Session Init can read its Job Configuration files from multiple directories. The list of directories jobs are read from is as follows (in order):

  • $XDG_CONFIG_HOME/upstart/ (or $HOME/.config/upstart/ if $XDG_CONFIG_HOME not set).
  • $HOME/.init/ (deprecated - supported for legacy User Jobs).
  • /usr/share/upstart/sessions/

The name of each job is taken to be the basename when any of the directory names above have been removed. For example, if a job configuration file exists as $HOME/.config/upstart/hello/world.conf, its name will be "hello/world" whereas if a job configuration file exists as /usr/share/upstart/sessions/foo/bar.conf, its name will be "foo/bar".

Upstart resolves any name collisions by simply accepting the first valid job (or override file) that it finds. For example, if the following two file exist:


Only the first, $HOME/.init/foo.conf will be used. Whereas if the following files exist:


Upstart will first read $HOME/.init/foo.conf, and then apply any changes in $HOME/.config/upstart/foo.override.

4.2.4   Odd Jobs   Job with start on, but no stop on

A job does not necessarily need a stop on stanza. If it lacks one, any running instances can still be stopped by an Administrator running either of:

  • initctl stop <job>
  • stop <job>

However, if such a job is not stopped, it may be stopped either by another job, or some other facility [32]. Worst case, if nothing else stops it, all processes will obviously be killed when the system is powered off.   Job with stop on, but no start on

If a job has no start on stanza, it can only be started manually by an Administrator running either of:

  • initctl start <job>
  • start <job>

If any job instances are running at system shutdown time, Upstart will stop them.   Minimal Job Configuration

What is the minimum content of a job configuration file? Interestingly enough, to be valid a job configuration file:

  • must not be empty
  • must be syntactically correct
  • must contain at least one legal stanza

Therefore, some examples of minimal job configuration files are:

  • Comments only:

    # this is an abstract job containing only a comment
  • author stanza only:

    author "foo"
  • description stanza only:

    description "this is an abstract job"

As shown, these are all example of Abstract Job configuration files.

4.3   Session Init

As of Upstart v1.7, Upstart has the ability to run as a non-PID 1 process (see upstart-user-sessions-spec for full details).

But why would you want to run another instance of Upstart? Well, due to its elegant design which assumes a dynamic system, it is perfectly suited to managing a users session. Traditionally, this job has been handled by applications such as "gnome-session", but by moving to an Upstart-based design a lot of benefits come "for free":

  • Event-based desktop.
  • Start desktop elements as-and-when needed or appropriate.
  • All the power of Upstarts process supervision.
  • Automatic logging of every job that runs. See console log.

To run a Session Init, simply arrange for the first process that starts a session to be run as "init --user". As when running as PID 1, the Session Init will emit the "startup" event that jobs can use to react to. All jobs that are managed by a Session Init have their parent set to the Session Init, not the system init. This is because a Session Init process is a true "sub-init". Jobs are loaded from potentially multiple directories. See Session Job for details.

The advent of Session Inits removes all need for User Jobs. These continue to be supported since the Session Init still reads the job configuration files from the User Job directory, but that directory is deprecated. See Session Job for further details.

4.3.1   Non-graphical Sessions ()

As of Ubuntu Saucy Salamander (13.10), a Session Init is used to manage the default graphical user session.

However, what if you want to use a Session Init on a server? This is not fully supported right now, but can be achieved as follows.

Create two System Job similar to the following...

  • session-init-setup.conf:

     start on runlevel [2345]
     stop on runlevel [!2345]
     # XXX: configurable
     env user=james
     export user
       uid=$(getent passwd "$user"|cut -d: -f3)
       gid=$(getent passwd "$user"|cut -d: -f4)
       # Create directory that would normally be
       # created by PAM when a user logs in.
       export XDG_RUNTIME_DIR="/run/user/$uid"
       mkdir -p "$XDG_RUNTIME_DIR"
       chmod 0700 "$XDG_RUNTIME_DIR"
       chown "$uid:$gid" "$XDG_RUNTIME_DIR"
      start session_init USER="$user"
    end script
  • session-init.conf:

    instance $USER
    stop on runlevel [016]
      uid=$(getent passwd "$USER"|cut -d: -f3)
      HOME=$(getent passwd "$USER"|cut -d: -f6)
      export XDG_RUNTIME_DIR="/run/user/$uid"
      export HOME
      exec su -s /bin/sh -c 'exec "$0" "$@"' $USER -- init --user
    end script


  • Two jobs are required since the setuid and setgid stanzas:
    • Apply to all job processes. Hence, it is not possible for a single job running as a non-privileged user to create the directory tree required by $XDG_RUNTIME_DIR.
    • Do not currently expand variables.
  • It is imperative to set $HOME and $XDG_RUNTIME_DIR in the Session Inits environment:
    • $HOME is needed by most user applications.
    • $XDG_RUNTIME_DIR is needed by the Session Init.

The session-init-setup job will start when the system is in a suitable state (disks mounted writeable and networking up). That job will start the session-init instance job which will start the actual Session Init (which will read Job Configuration Files from the usual locations for a Session Init).

To start a Session Init manually:

$ start session-init-setup
session-init-setup stop/waiting
$ status session-init USER=james
session-init (james) start/running, process 2442

Note that it is possible to specify that only certain Job Configuration File directories are read for a Session Init by specifying the --confdir option multiple times. For example:

init --user --confdir /etc/james/ --confdir /etc/bob/

Now, the Session Init will only read Job Configuration Files from /etc/james/ and /etc/bob/.

Note that this behaviour is Session Init-specific: without --user, the system Upstart would read Job Configuration Files from the /etc/bob/ directory only.   Joining a Session

If you have multiple sessions running for a user, or have started a Session Init from a System Job as shown in the example above, it is possible to "join" the appropriate session by simply setting the $UPSTART_SESSION environment variable.

For example:



$ export XDG_RUNTIME_DIR=/run/user/$(id -u)
$ initctl list-sessions
2983 unix:abstract=/com/ubuntu/upstart-session/1000/2983
$ export UPSTART_SESSION=unix:abstract=/com/ubuntu/upstart-session/1000/2983
$ initctl list
dbus start/running, process 3188
upstart-file-bridge start/running, process 3339
gnome-settings-daemon start/running, process 3206
re-exec stop/waiting
upstart-event-bridge start/running, process 3192

The initctl list command above will now list jobs in the users session specified by the $UPSTART_SESSION environment variable.

4.4   Event

A notification is sent by Upstart to all interested parties (either jobs or other events). Events can generally be thought of as "signals", "methods", or "hooks" [25], depending on how they are emitted and/or consumed.

Events are emitted (created and then broadcast) to the entire Upstart system. Note that it is not possible to stop any other job or event from seeing an event when it is emitted.

If there are no jobs which have registered an interest in an event in either their start on or stop on conditions, the event has no effect on the system.

Events can be created by an administrator at any time using:

# initctl emit <event>

Note that some events are "special". See the upstart-events(7) manual page for a list.

Note also that an event name with the same name as a job is allowed.

Jobs are often started or stopped as a result of other jobs starting or stopping. Upstart has a special set of events that it emits to announce these job state transitions. You'll probably notice that these events have the same names as some of the job states described in Job States, however it's important to appreciate that these are not describing the same thing. Task states are not events, and events are not task states. See Events, not States for details.

These events are as follows:

This event is emitted by Upstart when a job has been scheduled to run and is about to start executing.
This event is emitted by Upstart when a job is now running. Note that a job does not have to have an associated program or script so "running" does not necessarily imply that any additional process is executing.
This event is emitted by Upstart when a job is about to be stopped.
This event is emitted by Upstart when a job has completed (successfully or otherwise).

See Job Lifecycle for further details.

To help reinforce the difference, consider how Upstart itself starts: See the Startup Process.

  1. It performs its internal initialization.
  2. Upstart itself emits a single event called startup(7). This event triggers the rest of the system to initialize. Note that there is no "startup" job (and hence no /etc/init/startup.conf file).
  3. init(8) runs the mountall job (as defined in /etc/init/mountall.conf) since the startup(7) event satisfies mountall(8)'s requirement: "start on startup".
  4. The mountall(8) job in turn emits a number of events (including local-filesystems(7) and all-swaps(7)). See upstart-events(7) for further details.

Upstart provides three different types of Events.

4.4.1   Event Types   Signals

A Signal Event is a non-blocking (or asynchronous) event. Emitting an event of this type returns immediately, allowing the caller to continue. Quoting from [26]:

The announcer of a signal cares not whether anybody cared about it, and doesn't wait around to see whether anything happened. As far as the announcer cares, it's informational only.

Signal Events are created using the --no-wait option to the initctl emit command like this:

# initctl emit --no-wait mysignal

The non-blocking behaviour directly affects the emitter by allowing it to continue processing without having to wait for any jobs which make use of the event. Jobs which make use of the event (via start on or stop on) are also affected, as they're unable to stop, delay, or in any other way "hold up" the operation of the emitter.   Methods

A Method Event is a blocking (or synchronous) event which is usually coupled with a task. It acts like a method or function call in programming languages in that the caller is requesting that some work be done. The caller waits for the work to be done, and if problems were encountered, it expects to be informed of this fact.

Emitting a Method Event is simple:

# initctl emit mymethod

This is exactly like a Signal Event, except the event is being emitted synchronously such that the emitter has to wait until the initctl command completes. Once the initctl command has completed, there are two possible outcomes for the task that starts on Event mymethod:

  • The task runs successfully.
  • The task failed for some reason.

Assuming we have a job configuration file /etc/init/myapp.conf like this:

start on mymethod
exec /usr/bin/myapp $ACTION

You could start the myapp job and check if the "method" worked as follows:

# initctl emit mymethod ACTION=do_something
[ $? -ne 0 ] && { echo "ERROR: myapp failed"; exit 1; }   Hooks

A Hook Event is a blocking (or synchronous) event. Quoting from [27]:

"A hook is somewhere between a signal and a method. It's a notification that something changed on the system, but unlike a signal, the emitter waits for it to complete before carrying on."

Hooks are therefore used to flag to all interested parties that something is about to happen.

The canonical examples of Hooks are the two job events starting(7) and stopping(7), emitted by Upstart to indicate that a job is about to start and about to stop respectively.

4.4.2   Events, not States

Although Upstart does use states internally (and these are exposed via the list and status commands in initctl(8)), events are the way that job configuration files specify the desired behaviour of jobs: starting(7), started(7), stopping(7), stopped(7) are events, not states. These events are emitted "just prior" to the particular transition occurring. For example, the starting(7) event is emitted just before the job associated with this event is actually queued for start by Upstart.

4.5   Job Lifecycle

4.5.1   Starting a Job

  1. Initially the job is "at rest" with a goal of stop and a state of waiting (shown as stop/waiting by the initctl list and initctl status commands).

  2. The goal is changed from stop to start indicating the job is attempting to start.

  3. The state is changed from waiting to starting.

  4. The starting(7) event is emitted denoting the job is "about to start".

  5. Any jobs whose start on (or stop on) condition would be satisfied by this job starting are started (or stopped respectively).

  6. The starting(7) event completes.

  7. The state is changed from starting to pre-start.

  8. If the pre-start stanza exists, the pre-start process is spawned.

  9. If the pre-start process fails, the goal is changed from start to stop, and the stopping(7) and stopped(7) events are emitted with appropriate variables set denoting the error.

  10. Assuming the pre-start did not fail or did not call "stop", the main process is spawned.

  11. The state is changed from pre-start to spawned.

  12. Upstart then ascertains the final PID for the job which may be a descendent of the immediate child process if expect fork or expect daemon has been specified.

  13. The state is changed from spawned to post-start.

  14. If the post-start stanza exists, the post-start process is spawned.

  15. The state is changed from post-start to running.

  16. The started(7) event is emitted.

    For services, when this event completes the main process will now be fully running. If the job refers to a task, it will now have completed (successfully or other‐wise).

  17. Any jobs whose start on (or stop on) condition would be satisfied by this job being started are started (or stopped respectively).

4.5.2   Stopping a Job

  1. Assuming the job is fully running, it will have a goal of start and a state of running (shown as start/running by the initctl list and initctl status commands).

  2. The goal is changed from start to stop indicating the job is attempting to stop.

  3. The state is changed from running to pre-stop.

  4. If the pre-stop stanza exists, the pre-stop process is spawned.

  5. The state is changed from pre-stop to stopping.

  6. The stopping(7) event is emitted.

    The stopping event has a number of associated environment variables:

    • JOB

      The name of the job this event refers to.


      The name of the instance of the job this event refers to. This will be empty for single-instance jobs (those jobs that have not specified the instance stanza).

    • RESULT

      This variable will have the value "ok" if the job exited normally or "failed" if the job exited due to failure. Note that Upstart's view of success and failure can be modified using the normal exit stanza.


      The name of the script section that resulted in the failure. This variable is not set if RESULT=ok. If set, the variable will have one of the following values:

      • pre-start
      • post-start
      • main (denoting the script or exec stanza)
      • pre-stop
      • post-stop
      • respawn (denoting the job attempted to exceed its respawn limit)

      Either EXIT_STATUS or EXIT_SIGNAL will be set, depending on whether the job exited itself (EXIT_STATUS) or was stopped as a result of a signal (EXIT_SIGNAL).

      If neither variable is set, the process in question failed to spawn (for example, because the specified command to run was not found).

  7. Any jobs whose start on (or stop on) condition would be satisfied by this job stopping are started (or stopped respectively).

  8. The main process is stopped:

    • The signal specified by the kill signal stanza is sent to the process group of the main process. (such that all processes belonging to the jobs main process are killed). By default this signal is SIGTERM.

      See signal(7) and init(5).

    • Upstart waits for up to kill timeout seconds (default 5 seconds) for the process to end.

    • If the process is still running after the timeout, a SIGKILL signal is sent to the process which cannot be ignored and will forcibly stop the processes in the process group.

  9. The state is changed from killed to post-stop.

  10. If the post-stop stanza exists, the post-stop process is spawned.

  11. The state is changed from post-stop to waiting.

  12. The stopped(7) event is emitted.

    When this event completes, the job is fully stopped.

  13. Any jobs whose start on (or stop on) condition would be satisfied by this job being stopped are started (or stopped respectively).

Note: this information is also available in upstart-events(7).

4.6   Ordering

4.6.1   Order in which Events are Emitted

As a general rule, you cannot rely upon the the order in which events will be emitted. Your system is dynamic and Upstart responds to changes as-and-when they occur (for example hot-plug events).

That said, most systems which use Upstart provide a number of "well-known" events which you can rely upon.

For example on Ubuntu, these are documented in the upstart-events(7) man page, which is included within this document for convenience in appendix Ubuntu Well-Known Events (ubuntu-specific).

4.6.2   Order in Which Jobs Which start on the Same Event are Run

Assume you have three jobs like this:

  • /etc/init/X.conf

    start on event-A
  • /etc/init/Y.conf

    start on event-A
  • /etc/init/Z.conf

    start on event-A

Question: If event event-A is emitted, which job will run first?

Answer: It is not possible to say, and indeed you should not make any assumptions about the order in which jobs with the same conditions run in.

4.6.3   Ordering of Stop/Start Operations   Single Job

Imagine a job configuration file /etc/init/odd.conf like this:

start on event-A
stop  on event-A

  sleep 999
end script

Would Upstart be happy with this? Actually, yes it would! Upstart always handles stop on stanzas before handling start on stanzas. This means that this strange job would first be stopped (if it's currently running), then it would be started.

We can see what happens when we run this job more clearly when we increase the log priority to debug (see Change the log-priority):

# initctl log-priority debug

Now, we can watch the state transitions by viewing the system log.   If Job is Not Currently Running
# status odd
odd stop/waiting
# initctl emit event-A
# status odd
odd start/running, process 9474

And here is an example from the system log (with annotations) showing what happened:

event_new: Pending event-A event               # Upstart emitted the event.
Handling event-A event
event_pending_handle_jobs: New instance odd    # Job instance created.
odd goal changed from stop to start            # Since job not running,
odd state changed from waiting to starting     # change goal to "start".
event_new: Pending starting event
Handling starting event
event_finished: Finished starting event
odd state changed from starting to pre-start
odd state changed from pre-start to spawned
odd main process (9474)                        # Start script section.
odd state changed from spawned to post-start
odd state changed from post-start to running   # Job now fully started.
event_new: Pending started event
Handling started event
event_finished: Finished started event
event_finished: Finished event-A event   If Job is Currently Running
# status odd
odd stop/waiting
# start odd
odd start/running, process 11416    # Note this PID!
# status odd
odd start/running, process 11416
# initctl emit event-A
# status odd
odd start/running, process 11428    # Look! It changed!

Here is an example from the system log showing what happened in more detail. First the entries relating to starting the job:

odd goal changed from stop to start
odd state changed from waiting to starting
event_new: Pending starting event
Handling starting event
event_finished: Finished starting event
odd state changed from starting to pre-start
odd state changed from pre-start to spawned
odd main process (11416)
odd state changed from spawned to post-start
odd state changed from post-start to running
event_new: Pending started event
Handling started event
event_finished: Finished started event

Now, the event is emitted:

event_new: Pending event-A event
Handling event-A event
odd goal changed from start to stop             # Job already running, so stop it.
odd state changed from running to pre-stop
odd state changed from pre-stop to stopping
event_new: Pending stopping event
event_pending_handle_jobs: New instance odd
odd goal changed from stop to start
Handling stopping event
event_finished: Finished stopping event
odd state changed from stopping to killed
Sending TERM signal to odd main process (11416) # Forcibly stop existing job process.
odd main process (11416) killed by TERM signal  # Successfully stopped it.
odd state changed from killed to post-stop
odd state changed from post-stop to starting
event_new: Pending starting event
Handling starting event
event_finished: Finished starting event
odd state changed from starting to pre-start
odd state changed from pre-start to spawned
odd main process (11428)                        # New instance of job started with new PID.
odd state changed from spawned to post-start
odd state changed from post-start to running
event_new: Pending started event
Handling started event
event_finished: Finished started event
event_finished: Finished event-A event   Multiple Jobs

Upstart guarantees that jobs which stop on a particular event are processed before jobs that start on the same event.

Consider two jobs like this:

  • A.conf:

    start on startup
    stop on foo
  • B.conf:

    start on foo

Assuming that job "A" is already running, if the "foo" event is emitted, Upstart will always stop job "A" before starting job "B".

4.7   Runlevels

A runlevel is a single-byte name for a particular system configuration. Runlevels for Debian and Ubuntu systems are generally as follows [34]:

  • 0 : System halt.
  • 1 : Single-User mode.
  • 2 : Graphical multi-user plus networking (DEFAULT)
  • 3 : Same as "2", but not used.
  • 4 : Same as "2", but not used.
  • 5 : Same as "2", but not used.
  • 6 : System reboot.

There are also a few pseudo-runlevels:

  • N : The previous runlevel cannot be determined.
  • S : Alias for Single-User mode.

4.7.1   Display Runlevel

To display your current and previous runlevels separated by a space character, run the /sbin/runlevel command. Note that if this command is unable to determine the system runlevel, it may display simply "unknown":

$ runlevel
N 2

The output above shows that:

  • there was no previous runlevel (the system was booted and went straight to the current runlevel).
  • the current runlevel is "2".

4.7.2   Change Runlevel Immediately

To change runlevel immediately, use one of the commands below:

4.7.3   Changing the Default Runlevel   Permanently

To change the default runlevel the system will boot into, modify the variable DEFAULT_RUNLEVEL in file /etc/init/rc-sysinit.conf. For example, to make the system boot by default to single user mode, set:

env DEFAULT_RUNLEVEL=1   Single Boot

If you want to change the default runlevel for a single boot, rather than making the change permanent by modify the rc-sysinit.conf file, simply append the variable to the kernel command line:


Traditionally, the default runlevel was encoded in file /etc/inittab. However, with Upstart, this file is no longer used (it is supported by Upstart, but its use is deprecated).

5   System Phases

The information in this section relates to an Ubuntu system.

To obtain a better understanding of how jobs and events relate at startup and shutdown time, see Visualising Jobs and Events.

5.1   Startup

At boot, after the initramfs system has been run (for setting up RAID, unlocking encrypted file system volumes, et cetera), Upstart will be given control. The initramfs environment will exec(3) /sbin/init (this is the main Upstart binary) and cause it to run as PID 1.

5.1.1   Startup Process

Note that in this section we assume the default runlevel is "2". See Changing the Default Runlevel for further details.

  1. Upstart performs its internal initialization.

  2. Upstart itself emits a single event called startup(7).

    This event triggers the rest of the system to initialize [33].

  3. init(8) runs a small number of jobs which specify the startup(7) event in their start on condition.

    The most notable of these is the mountall job which mounts your disks and filesystems.

  4. The mountall(8) job in turn emits a number of events.

    These include local-filesystems(7), virtual-filesystems(7) and all-swaps(7). See upstart-events(7) for further details.

  5. The virtual-filesystems(7) event causes the udev job to start.

  6. The udev job causes the upstart-udev-bridge job to start.

  7. The upstart-udev-bridge job will at some point emit the "net-device-up IFACE=lo" event signifying the local network (for example, for IPv4) is available.

  8. After the last filesystem is mounted, mountall(8) will emit the filesystem event.

  9. Since the start on condition for the rc-sysinit job is:

    start on filesystem and net-device-up IFACE=lo

    Upstart will then start the rc-sysinit job.

  10. The rc-sysinit job calls the telinit command, passing it the runlevel to move to:

    telinit 2
  11. The telinit command emits the runlevel(7) event as:


    Note that this is all the telinit command does – it runs no commands itself to change runlevel!

    See Runlevels for further information on runlevels.

  12. The runlevel(7) event causes many other Upstart jobs to start, including /etc/init/rc.conf which starts the legacy SystemV init system.

5.2   Shutdown

5.2.1   Observations

There are some important points related to system shutdown:

  • Upstart never shuts down itself

    Upstart will "die" when the system is powered off, but if it ever exits, that is a bug.

  • Upstart never stops a job with no stop on condition.

  • Ubuntu employs both Upstart and SysV jobs.

    Ubuntu currently employs a hybrid system where core services are handled by Upstart, but additional services can be run in the legacy SystemV mode. This may seem odd, but consider that there are thousands of packages available in Ubuntu via the Universe and Multiverse repositories and hundreds of services. To avoid having to change every package to work with Upstart, Upstart allows packages to utilize their existing SystemV (and thus Debian-compatible) scripts.

5.2.2   Shutdown Process

To initiate a shutdown, perform one of the following actions:

  • Click "Shut Down..." (or equivalent) in your graphical environment (for example Gnome)

  • Run the shutdown(8) command, for example:

    # shutdown -h now

The following steps will now be taken:

  1. Assuming the current runlevel is "2", either of the actions above will cause Upstart to emit the runlevel(7) event like this:

    runlevel RUNLEVEL=0 PREVLEVEL=2
  2. The job /etc/init/rc.conf will be run.

    This job calls /etc/init.d/rc passing it the new runlevel ("0").

  3. The SystemV system will then invoke the necessary scripts in /etc/rc0.d/ to stop SystemV services.

  4. One of the scripts run is /etc/init.d/sendsigs.

    This script will kill any remaining processes not already stopped (including Upstart processes).

5.3   Reboot

To initiate a reboot, perform one of the following actions:

  • Click "Restart..." (or equivalent) in your graphical environment (for example Gnome)

  • Run the shutdown(8) command specifying the "-r" option, for example:

    # shutdown -r now
  • Run the reboot(8) command:

    # reboot

The following will steps will now be taken:

  1. Assuming the current runlevel is "2", whichever command is run above will cause Upstart to emit the runlevel(7) event like this:

    runlevel RUNLEVEL=6 PREVLEVEL=2
  2. The job /etc/init/rc.conf will be run.

    This job calls /etc/init.d/rc passing it the new runlevel ("6").

  3. The SystemV system will then invoke the necessary scripts in /etc/rc6.d/ to stop SystemV services.

  4. One of the scripts run is /etc/init.d/sendsigs.

    This script will kill any remaining processes not already stopped (including Upstart processes).

5.4   Single-User Mode

When booting direct into single-user mode, the runlevel command will show:

# runlevel

See Runlevels.

5.5   Recovery Mode ()

Ubuntu provides a recovery mode in case your system experiences problems. This is handled by the friendly-recovery package. If you select a "recovery mode" option on the Grub menu. This makes the initramfs pass a flag to Upstart which ensures that the /etc/init/friendly-recovery.conf Upstart job is the first job run after Upstart starts. As a result, this job has full control over the system and provides a friendly menu that allows users to check disks with fsck(8), repair your package database and so on.

5.6   Failsafe Mode ()

This is a new phase introduced in Ubuntu 11.10 that borrows an idea from Google's Chrome OS. A new job called failsafe has been introduced that checks to ensure the system has reached a particular state. If the expected state is not attained, the job reboots the system automatically.

6   Configuration

This section lists a number of job configuration file stanzas, giving example usage for each. The reference for your specific version of Upstart will be available in the init(5) man page. [18]

6.1   Stanzas by Category

Configuration Stanzas by Category (detail in brackets show version of Upstart stanza added)
Category Stanzas Added in Version
Process Definition exec  
Event Definition manual 0.6.7
start on  
stop on  
Job Environment env  
Services, tasks and respawning normal exit  
respawn limit  
Instances instance  
Documentation author  
usage 1.5
Process environment apparmor load 1.9
apparmor switch 1.9
cgroup 1.13
console none  
console log 1.4
console output  
console owner  
oom score  
setgid 1.4
setuid 1.4
Process Control expect fork  
expect daemon  
expect stop  
kill signal 1.3
kill timeout  
reload signal 1.10

6.2   apparmor

6.2.1   apparmor load

Load specified AppArmor Mandatory Access Control system profile into the kernel prior to starting the job. The main job process (as specified by exec or script) will be confined to this profile.


apparmor load <profile-path>


  • <profile-path> must be an absolute path.
  • The job will fail if the profile doesn't exist, or the profile fails to load.


apparmor load /etc/apparmor.d/usr.sbin.cupsd
exec /usr/sbin/cupsd -F

6.2.2   apparmor switch

Run main job process with already-loaded AppArmor Mandatory Access Control system profile.


apparmor switch <profile-name>


  • The job will fail if the profile named does not exist, or is not already loaded.


apparmor switch /usr/bin/cupsd
exec /usr/sbin/cupsd -F

6.3   author


author <string>

Quoted name (and maybe contact details) of author of this Job Configuration File.


author "Scott James Remnant <>"

6.4   cgroup

Upstart 1.13 supports cgroups with the aid of cgmanager (see cgmanager(8)).

A new "cgroup" stanza is introduced that allows job processes to be run within the specified cgroup.



This allows the job to specify the control group all job processes will run in and optionally specify a setting for the particular cgroup.


  • This stanza will be ignored if the version of Upstart is new enough to support cgroups but has been built without cgroup support.
  • This stanza will also be ignored if the init(8) daemon has had cgroup support disabled at boot time (see init(8)).
  • A job which specifies this stanza will not be started until both its start on condition is met and the address of the cgroup manager has been communicated to the init(8) daemon using the initctl(8) command initctl notify-cgroup-manager-address.

If only the cgroup controller (such as memory, cpuset, blkio) is specified, a job-specific cgroup will be created and the job processes placed in it. The form of this cgroup is:


... or if the job specifies the instance stanza the group will be the expanded value of:


Any forward slashes in $UPSTART_JOB and $UPSTART_INSTANCE will be replaced with underscore ("_") characters.

This default cgroup for the job may be specified explicitly within a NAME using the special variable "$UPSTART_CGROUP". This variable is not an environment variable and is only valid within the context of the cgroup stanza.

If NAME is not specified or does not contain "$UPSTART_CGROUP", the job processes will not be placed in an upstart-specific group.

Note that this special variable cannot be specified with enclosing braces around the name.

No validation is performed on the specified values until the job is due to be started.

If the CONTROLLER is invalid, or the NAME cannot be created or the KEY or VALUE are invalid, the job will be failed.

The NAME argument may contain any valid variable and can also contain forward slashes to run the job processes in a sub-cgroup.

If any argument contains space characters, it must be quoted.

If a KEY is specified, a VALUE must also be specified (even it is simply an empty string).

The stanza maybe specified multiple times. The last occurence will be used except in the scenario where each occurence specifies a different KEY in which case all the keys and values will be applied.

It is not an error if NAME already exists.

Valid syntax examples:

  • Implicit NAME, no setting:

    cgroup CONTROLLER
  • Explicit NAME, no setting:

  • Implicit NAME with setting:

  • Explicit NAME with setting:



  • Run all job processes in the default cpu cgroup controller group:

    cgroup cpu
  • As above:

    cgroup cpu $UPSTART_CGROUP
  • As above:

    cgroup cpu "$UPSTART_CGROUP"
  • Attempt to place the job processes in a non-job-specific cgroup:

    cgroup cpu "a-well-known-cgroup"
  • The job will only start once the manager is up and running and will have a 50MB memory limit, be restricted to CPU ids 0 and 1 and have a 1MB/s write limit to the block device 8:16. The job will fail to start if the system has less than 50MB of RAM or less than 2 CPUs:

    cgroup memory $UPSTART_CGROUP limit_in_bytes 52428800
    cgroup cpuset $UPSTART_CGROUP cpus 0-1
    cgroup blkio slowio throttle.write_bps_device "8:16 1048576"

6.5   console

For all versions of Upstart prior to v1.4, the default value for console was console none. As of Upstart 1.4, the default value is console log. If you are using Upstart 1.4 or later and wish to retain the old default, boot specifying the --no-log command-line option. An alternative is to boot using the --default-console <value> option which allows the default console value for jobs to be specified. Using this option it is possible to set the default to none but still honour jobs that specify explicitly console log.

6.5.1   console log

Connects standard input to /dev/null. Standard output and standard error are connected to one end of a pseudo-terminal such that any job output is automatically logged to a file in directory /var/log/upstart/ for System Jobs and $XDG_CACHE_HOME/upstart/ (or $HOME/.cache/upstart/ if $XDG_CACHE_HOME is not set) for Session Jobs.

The log directory can be changed by specifying the --logdir <directory> command-line option.

If a User Job running in a pre-Upstart 1.7 environment specifies this stanza, Upstart will treat the job as if it had specified console none.

6.5.2   console none

Connects the job's standard input, standard output and standard error file descriptors to /dev/null.

6.5.3   console output

Connects the job's standard input, standard output and standard error file descriptors to the console device.   Example of console output

console output

pre-start script

  # Perform whatever checks you like here (maybe checking
  # '/etc/default/foo' to see if the service is enabled # or not).
  # if there are no problems detected, simply "exit 0", else do
  # something like this...

  # display an error message to stderr *on the console* and also write
  # the same message to the system log.
  logger -is -t "$UPSTART_JOB" "ERROR: foo!"

  # tell Upstart not to start the main process for the job.
  exit 1
end script

# this service doesn't do much :-)
exec sleep 999

See pre-start.

6.5.4   console owner

Identical to console output except that additionally it makes the job the owner of the console device. This means it will receive certain signals from the kernel when special key combinations such as Control-C are pressed.

6.6   chdir


chdir <directory>

Runs the job's processes with a working directory in the specified directory instead of the root of the filesystem.


chdir /var/mydaemon

6.7   chroot


chroot <directory>

Runs the job's processes in a chroot(8) environment underneath the specified directory.

Note that the specified directory must have all the necessary system libraries for the process to be run, often including /bin/sh.


chroot /srv/chroots/oneiric

6.8   description


description <string>

One line quoted description of Job Configuration File. For example:

description "OpenSSH server"

6.9   emits


emits <values>

Specifies the events the job configuration file generates (directly or indirectly via a child process). This stanza can be specified multiple times for each event emitted. This stanza can also use the following shell wildcard meta-characters to simplify the specification:

  • asterisk ("*")
  • question mark ("?")
  • square brackets ("[" and "]")

For example, upstart-udev-bridge can emit a large number of events. Rather than having to specify every possible event, since the form of the event names is consistent, a single emits stanza can be specified to cover all possible events:

emits *-device-*

Further Examples:

emits foo-event bar-event wibble-event
emits hello

6.10   end script

This pseudo-stanza acts as a terminator for script sections:

6.11   env



Allows an environment variable to be set which is accessible in all script sections.


env myvar="hello world"

  echo "myvar='$myvar'" > /run/script.log
end script

See Environment Variables.

6.12   exec


exec COMMAND [ ARG ]...

Stanza that allows the specification of a single-line command to run. Note that if this command-line contains any shell meta-characters, it will be passed through a shell prior to being executed. This ensures that shell redirection and variable expansion occur as expected.


exec /usr/bin/my-daemon --option foo -v

6.13   expect


This stanza is extremely important: read this section carefully!

Upstart will keep track of the process ID that it thinks belongs to a job. If a job has specified the instance stanza, Upstart will track the PIDs for each unique instance of that job.

If you do not specify the expect stanza, Upstart will track the life cycle of the first PID that it executes in the exec or script stanzas. However, most Unix services will "daemonize", meaning that they will create a new process (using fork(2)) which is a child of the initial process. Often services will "double fork" to ensure they have no association whatsoever with the initial process. (Note that no services will fork more than twice initially since there is no additional benefit in doing so).

In this case, Upstart must have a way to track it, so you can use expect fork, or expect daemon which allows Upstart to use ptrace(2) to "count forks".

To allow Upstart to determine the final process ID for a job, it needs to know how many times that process will call fork(2). Upstart itself cannot know the answer to this question since once a daemon is running, it could then fork a number of "worker" processes which could themselves fork any number of times. Upstart cannot be expected to know which PID is the "master" in this case, considering it does not know if worker processes will be created at all, let alone how many times, or how many times the process will fork initially. As such, it is necessary to tell Upstart which PID is the "master" or parent PID. This is achieved using the expect stanza.

The syntax is simple, but you do need to know how many times your service forks.

Note that most daemons fork twice.

If your daemon has a "don't daemonize" or "run in the foreground" mode, then it's much simpler to use that and not run with fork following. One issue with that though, is that Upstart will emit the started JOB=yourjob event as soon as it has executed your daemon, which may be before it has had time to listen for incoming connections or fully initialize.

A final point: the expect stanza only applies to exec and script stanzas: it has no effect on pre-start and post-start.

It's important to note that the "expect" stanza is thus being used for two different but complementary tasks:

  • Identifying service readiness.
  • PID tracking.

6.13.1   expect fork

Upstart will expect the process executed to call fork(2) exactly once.

Some daemons fork a new copy of themselves on SIGHUP, which means when the Upstart reload command is used, Upstart will lose track of this daemon. In this case, expect fork cannot be used. See Daemon Behaviour.

6.13.2   expect daemon

Upstart will expect the process executed to call fork(2) exactly twice.

6.13.3   expect stop

Specifies that the job's main process will raise the SIGSTOP signal to indicate that it is ready. init(8) will wait for this signal and then:

  1. Immediately send the process SIGCONT to allow it to continue.
  2. Run the job's post-start script (if any).

Only then will Upstart consider the job to be running.

6.13.4   How to Establish Fork Count

If the application you are attempting to create a Job Configuration File does not document how many times it forks, you can run it with a tool such as strace(1) which will allow you to count the number of forks. For example:

# Trace all children of /usr/bin/myapp
$ sudo strace -o /tmp/strace.log -fFv /usr/bin/myapp --arg foo --hello wibble &

# After allowing some "reasonable" time for the app to start, kill it and strace
$ sudo killall -9 strace

# Display the number of forks
#   1 => specify "expect fork"
#   2 => specify "expect daemon"
$ sudo egrep "\<(fork|clone)\>\(" /tmp/strace.log | wc | awk '{print $1}'

6.13.5   Implications of Misspecifying expect

The table below summarizes the behaviour resulting for every combination of expect stanza and number of fork(2) calls:

Expect Stanza Behaviour
  Specification of Expect Stanza
Forks no expect expect fork expect daemon
0 Correct start hangs start hangs
1 Wrong pid tracked † Correct start hangs
2 Wrong pid tracked † Wrong pid tracked † Correct


'' - No PID will be displayed.

6.13.6   Recovery on Misspecification of expect   When start hangs

The start command will "hang" if you have misspecified the expect stanza by telling Upstart to expect more fork(2) calls than your application actually makes.

To resolve the situation:

  1. Interrupt the start command by using "CONTROL+c" (or sending the process the SIGINT signal).

  2. Run the initctl status command for your job. You will see something like:

    myjob start/spawned, process 1234

    You'll notice that the PID shown is actually correct since Upstart has tracked the initial PID.

  3. Kill(1) the PID of your application.

  4. Re-run the initctl status command for your job. You will see something like:

    myjob stop/waiting
  5. Correct the expect stanza specification in the job configuration file.   When Wrong PID is Tracked

If you have misspecified the expect stanza by telling Upstart to expect fewer fork(2) calls than your application actually makes, Upstart will be unable to manage it since it will be looking at the wrong PID. The start command will start your job, but it will show unexpected output (the goal and state will be shown as stop/waiting).

To resolve the situation:

  1. Run the initctl status command for your job. You will see something like:

    myjob stop/waiting

    Notice that no PID is displayed.

  2. Find your jobs PID using ps(1). (If you're struggling to find it, remember that the parent PID will always be "1").

  3. Kill(1) the PID of your application.

  4. Correct the expect stanza specification in the job configuration file.

6.14   export

Export variables previously set with env to all events that result from this job. See for example Job Lifecycle.

Note that no leading dollar sign ($) is specified.


env myvar="hello world"
export myvar

6.15   instance

Sometimes you want to run the same job, but with different arguments. The variable that defines the unique instance of this job is defined with instance.

6.15.1   A Simple Instance Example

Let us start with a simple example which we will call "foo.conf":

instance $BAR

  . /etc/default/myapp-${BAR}

  echo "hello from instance $BAR"
  sleep 999
end script

The example above defines an instance job by specifying the instance stanza followed by the name of a variable (note that you MUST specify the dollar sign ('$').

Note that the entire job is the instance job: providing the instance stanza allows Upstart to make each running version of this job unique.

The job first sources an instance-specific configuration file ("myapp-${BAR}") then displays a message. Note again that we're now using that instance variable $BAR.

So, let's start an instance of this job:

$ sudo start foo
start: Unknown parameter: BAR

Oops! We forgot to specify the particular value for the BAR variable which makes each instance unique. Lets try again:

$ sudo start foo BAR=bar
foo (bar) start/running, process 1234

So, we now have one instance running. Let's start another:

$ sudo start foo BAR=bar
start: Job is already running: foo (bar)

Oops! We tried to run another instance with the same instance name (well, the same value of the BAR variable technically). Lets try again:

$ sudo start foo BAR=baz
foo (baz) start/running, process 1235

Okay. We should now have two instance running, but let us confirm that:

$ initctl list | grep ^foo
foo (bar) start/running, process 1234
foo (baz) start/running, process 1235

Good - Upstart is running two instances as expected. Notice the instance name in brackets after the job name in the initctl output above.

We will start one more instance:

$ sudo start foo BAR="hello world"
$ initctl list | grep ^foo
foo (bar) start/running, process 1234
foo (baz) start/running, process 1235
foo (hello world) start/running, process 1236

Let's try to stop the instances:

$ sudo stop foo
stop: Unknown parameter: BAR

That fails as Upstart needs to know which instance to stop and we didn't specify an instance value for the BAR instance variable. Rather than stopping each instance in turn, let's script it so that we can stop then all in one go:

$ initctl list | grep "^foo " | cut -d\( -f2 | cut -d\) -f1 | while read i
  sudo stop foo BAR="$i"
foo stop/waiting
foo stop/waiting
foo stop/waiting

All unique instances of the foo job are now stopped.

6.15.2   Another Instance Example

Lets say that once memcached is up and running, we want to start a queue worker for each directory in /var/lib/queues:

# queue-workers

start on started memcached


  for dir in `ls /var/lib/queues` ; do
    start queue-worker QUEUE=$dir
end script

And now:

# queue-worker

stop on stopping memcached


instance $QUEUE

exec /usr/local/bin/queue-worker $QUEUE

In this way, Upstart will keep them all running with the specified arguments, and stop them if memcached is ever stopped.

The instance stanza is designed to make a running job unique.


  • the stanza isn't restricted to a single value. You can do silly things like the following if you wish:

    instance ${myvar1}hello${myvar2}-foo/\wibble${var3}{$JOB}

    See Multiple Running Job Instances Without PID for another crazy real-life example.

  • You must include at least one variable and it must have a leading dollar sign ($):

    # GOOD (value can be changed by specifying different values
    # for the variable called 'foo')
    instance $foo
    # BAD (value will always be the string literal "foo")
    instance foo
  • If you attempt to start a job with the instance stanza, but forget to provide the required variables, you will get an error since Upstart cannot then guarantee uniqueness. For example, if you have a job configuration file foo.conf such as this:

    instance $bar
      sleep 999
    end script

    Attempting to start it without specifying a value for foo will fail:

    # start foo
    start: Unknown parameter: bar

    Let's try again:

    # start foo bar=1
    foo (1) start/running, process 30003

    And now let's start another instance:

    # start foo bar="hello 1,2,3"
    foo (hello 1,2,3) start/running, process 30008

    Finally, let's see the current state of our two job instances:

    $ initctl list|grep ^foo
    foo (1) start/running, process 30003
    foo (hello 1,2,3) start/running, process 30008

Note that to obtain correct restart behaviour, you would need to do something like the following:

# worker.conf
instance $id
exec myworker name=$id

# workers.conf
pre-start script
    for inst in a b c
        start worker id=$inst
end script

post-stop script
    for inst in `initctl list|grep "^worker "|awk '{print $2}'|tr -d ')'|tr -d '('`
        stop worker id=$inst
end script

Note that "workers.conf" has no main exec or script section - this "master" job will run (without a pid) for the duration that the slave or children (individual "worker") job instances run:

$ initctl list|grep ^worker
worker stop/waiting
workers stop/waiting
$ start workers
workers start/running
$ initctl list|grep ^worker
worker (c) start/running, process 12226
worker (b) start/running, process 12223
worker (a) start/running, process 12221
workers start/running
$ restart workers
workers start/running
$ initctl list|grep ^worker
worker (c) start/running, process 12246
worker (b) start/running, process 12244
worker (a) start/running, process 12242
workers start/running
$ stop workers
workers stop/waiting
$ initctl list|grep ^worker
worker stop/waiting
workers stop/waiting

Note further that if any worker fails to start or stop, this wil fail the overall "workers" job. If you don't want this behaviour, use the "|| true" trick:

# workers.conf
pre-start script
    for inst in a b c
        start worker id=$inst || :
end script

post-stop script
    for inst in `initctl list|grep "^worker "|awk '{print $2}'|tr -d ')'|tr -d '('`
        stop worker id=$inst || :
end script

6.15.3   Starting an Instance Job Without Specifying an Instance Value

Note that if you have a job which makes use of instance but which may need to be run manually by an administrator, it is possible to "cheat" and allow them to start the job without specifying an explicit instance value:

# /etc/init/trickery.conf
start on foo


Now, an Administrator can start this job as follows:

# start trickery

And this will work even if there is already a running instance of the trickery job (assuming the existing instance was started automatically).

This bit of trickery relies upon the fact that Upstart will set the $UPSTART_EVENTS environment variable before starting this job as a result of its start on condition becoming true. In this case, Upstart would therefore set UPSTART_EVENTS='foo'.

However, since the job sets a null default value for this variable, when an Administrator starts the job, UPSTART_EVENTS will be set to a null value. This empty value is enough to make that instance unique (since there are no other instances with a null instance value!)

See Environment Variables for details of $UPSTART_EVENTS.

6.16   kill signal

Specifies the stopping signal, SIGTERM by default, a job's main process will receive when stopping the running job.

The signal should be specified as a full name (for example SIGTERM) or a partial name (for example TERM). Note that it is possible to specify the signal as a number (for example 15) although this should be avoided if at all possible since signal numbers may differ between systems.


kill signal INT
kill signal SIGINT

Note that if you are running an older version of Upstart without this feature, and you have an application which breaks with the normal conventions for shutdown signal, you can simulate it to some degree by using start-stop-daemon(8) with the --signal option:

start on some-event

env cmd=/usr/bin/foo

exec start-stop-daemon --start --exec $cmd

pre-stop exec start-stop-daemon --signal QUIT --stop --exec $cmd

6.17   kill timeout

The number of seconds Upstart will wait before killing a process. The default is 5 seconds.


kill timeout 20

6.18   limit

Provides the ability to specify resource limits for a job.

For example, to allow a job to open any number of files, specify:

limit nofile unlimited unlimited


If a user job specifies this stanza, it may fail to start should it specify a value greater than the users privilege level allows.

For further details on the available limits see init(5) and getrlimit(2).

6.19   manual

Added in Upstart v0.6.7

This stanza will tell Upstart to ignore the start on / stop on stanzas. It is useful for keeping the logic and capability of a job on the system while not having it automatically start at boot-up.



6.20   nice

Change the jobs scheduling priority from the default. See nice(1).


# run with lowest priority
nice 19

6.21   normal exit

Used to change Upstart's idea of what a "normal" exit status is. Conventionally, processes exit with status 0 (zero) to denote success and non-zero to denote failure. If your application can exit with exit status 13 and you want Upstart to consider this as an normal (successful) exit, then you can specify:

normal exit 0 13

You can even specify signals. A signal can be specified either as a full name (for example SIGTERM) or a partial name (for example TERM);

For example, to consider exit codes 0 and 13 as success and also to consider the program to have completed successfully if it exits on signal SIGUSR1 and SIGWINCH, specify:

normal exit 0 13 SIGUSR1 SIGWINCH

Equivalently, you could specify:

normal exit 0 13 USR1 WINCH

6.22   oom score

Linux has an "Out of Memory" killer facility. This is a feature of the kernel that will detect if a process is consuming increasingly more memory. Once "triggered", the kernel automatically takes action by killing the rogue process to avoid it impacting the system adversely.

Normally the OOM killer regards all processes equally, this stanza advises the kernel to treat this job differently.

The "adjustment" value provided to this stanza may be an integer value from -999 (very unlikely to be killed by the OOM killer) up to 1000 (very likely to be killed by the OOM killer). It may also be the special value never to have the job ignored by the OOM killer entirely (potentially dangerous unless you really trust the application in all possible system scenarios).


# this application is a "resource hog"
oom score 1000

expect daemon
exec /usr/bin/leaky-app

6.23   post-start


post-start exec|script

Script or process to run after the main process has been spawned, but before the started(7) event has been emitted.

Use this stanza when a delay (or some arbitrary condition) must be satisfied before an executed job is considered "started". An example is MySQL. After executing it, it may need to perform recovery operations before accepting network traffic. Rather than start dependent services, you can have a post-start like this:

post-start script
  while ! mysqladmin ping localhost ; do sleep 1 ; done
end script

6.24   post-stop


post-stop exec|script

There are times where the cleanup done in pre-start is not enough. Ultimately, the cleanup should be done both pre-start and post-stop, to ensure the service starts with a consistent environment, and does not leave behind anything that it shouldn't.

exec /some/directory/script

If it is possible, you'll want to run your daemon with a simple exec line. Something like this:

exec /usr/bin/mysqld

If you need to do some scripting before starting the daemon, script works fine here. Here is one example of using a script stanza that may be non-obvious:

# statd - NSM status monitor

description   "NSM status monitor"
author                "Steve Langasek <>"

start on (started portmap or mounting TYPE=nfs)
stop on stopping portmap

expect fork

env DEFAULTFILE=/etc/default/nfs-common

pre-start script
    if [ -f "$DEFAULTFILE" ]; then
        . "$DEFAULTFILE"

    [ "x$NEED_STATD" != xno ] || { stop; exit 0; }

    start portmap || true
    status portmap | grep -q start/running
    exec sm-notify
end script

    if [ -f "$DEFAULTFILE" ]; then
        . "$DEFAULTFILE"

    if [ "x$NEED_STATD" != xno ]; then
        exec rpc.statd -L $STATDOPTS
end script

Because this job is marked respawn, an exit of 0 is "ok" and will not force a respawn (only exiting with a non-0 exit or being killed by an unexpected signal causes a respawn), this script stanza is used to start the optional daemon rpc.statd based on the defaults file. If NEED_STATD=no is in /etc/default/nfs-common, this job will run this snippet of script, and then the script will exit with 0 as its return code. Upstart will not respawn it, but just gracefully see that it has stopped on its own, and return to stopped status. If, however, rpc.statd had been run, it would stay in the start/running state and be tracked normally.

6.25   pre-start


pre-start exec|script

Use this stanza to prepare the environment for the job. Clearing out cache/tmp dirs is a good idea, but any heavy logic is discouraged, as Upstart job files should read like configuration files, not so much like complicated software.

pre-start script
  [ -d "/var/cache/squid" ] || squid -k
end script

Another possibility is to cancel the start of the job for some reason. One good reason is that it's clear from the system configuration that a service is not needed:

pre-start script
  if ! grep -q 'parent=foo' /etc/bar.conf ; then
    stop ; exit 0
end script

Note that the "stop" command did not receive any arguments. This is a shortcut available to jobs where the "stop" command will look at the current environment and determine that you mean to stop the current job.

6.25.1   pre-start example ( , )

On Ubuntu, the common pre-start idiom is to use /etc/default/myapp, so the example would become:

pre-start script

  # stop job from continuing if no config file found for daemon
  [ ! -f /etc/default/myapp ] && { stop; exit 0; }

  # source the config file
  . /etc/default/myapp

  # stop job from continuing if admin has not enabled service in
  # config file.
  [ -z "$ENABLED" ] && { stop; exit 0; }

end script

This is safe since the job will not start (technically it won't progress beyond the pre-start stage) if:

  • the config file does not exist.
  • the config file has not been modified to enable the service.

Note that the example above assumes your applications configuration file is shell-compatible (in other words it contains name="value" entries). If this is not the case, just use grep(1) or similar:

enabled=$(grep ENABLED=1 $CONFIG)
[ -z "$enabled" ] && exit 0

Or something like this:

if ! grep -q DISABLED=false /etc/default/myapp; then
  stop ; exit 0

See Example of console output for another of example where you can display an error message if the job detects it should not be started.

6.26   pre-stop


pre-stop exec|script

The pre-stop stanza will be executed before the job's stopping(7) event is emitted and before the main process is killed.

Stopping a job involves sending SIGTERM to it. If there is anything that needs to be done before SIGTERM, do it here. Arguably, services should handle SIGTERM very gracefully, so this shouldn't be necessary. However, if the service takes more than kill timeout seconds (default, 5 seconds) then it will be sent SIGKILL, so if there is anything critical, like a flush to disk, and raising kill timeout is not an option, pre-stop is not a bad place to do it. [20]

You can also use this stanza to cancel the stop, in a similar fashion to the way one can cancel the start in the pre-start.

6.27   reload signal

Specifies the signal that Upstart will send to the jobs main process when the job needs to be reloaded (the default is SIGHUP).

The signal should be specified as a full name (for example SIGHUP) or a partial name (for example HUP). Note that it is possible to specify the signal as a number (for example 1) although this should be avoided if at all possible since signal numbers may differ between systems.


reload signal SIGUSR1
reload signal USR1

6.28   respawn


If you are creating a new Job Configuration File, do not specify the respawn stanza until you are fully satisfied you have specified the expect stanza correctly. If you do, you will find the behaviour potentially very confusing.

Without this stanza, a job that exits quietly transitions into the stop/waiting state, no matter how it exited.

With this stanza, whenever the main script/exec exits, without the goal of the job having been changed to stop, the job will be started again. This includes running pre-start, post-start and post-stop. Note that pre-stop will not be run.

There are a number of reasons why you may or may not want to use this. For most traditional network services this makes good sense. If the tracked process exits for some reason that wasn't the administrator's intent, you probably want to start it back up again.

Likewise, for tasks, (see below), respawning means that you want that task to be retried until it exits with zero (0) as its exit code.

One situation where it may seem like respawn should be avoided, is when a daemon does not respond well to SIGTERM for stopping it. You may believe that you need to send the service its shutdown command without Upstart being involved, and therefore, you don't want to use respawn because Upstart will keep trying to start your service back up when you told it to shutdown.

However, the appropriate way to handle that situation is a pre-stop which runs this shutdown command. Since the job's goal will already be 'stop' when a pre-stop is run, you can shutdown the process through any means, and the process won't be re-spawned (even with the respawn stanza).

Note that if a job is respawned, the variable "$PROCESS" will be set to the name of the job process that failed (for example "pre-start" or "main"). See stopped(7) for further details.

Further note that if the job does not specify the respawn limit stanza as well as the respawn stanza, the job will have the default respawn limit applied (see respawn limit).

6.29   respawn limit

Yes, this is different to a plain respawn: specifying respawn limit does not imply respawn.


respawn limit COUNT INTERVAL | unlimited


# respawn the job up to 10 times within a 5 second period.
# If the job exceeds these values, it will be stopped and
# marked as failed.
respawn limit 10 5

# respawn the job indefinitely
respawn limit unlimited

Respawning is subject to a limit. If the job is respawned more than COUNT times in INTERVAL seconds, it will be considered to be having deeper problems and will be stopped. Default COUNT is 10. Default INTERVAL is 5 seconds.

To have the job respawn indefinitely, specify an argument of "unlimited". However, care should be taken using this option: does your service really stop that frequently? Should it?

Specifying either COUNT or INTERVAL as 0 (zero) implies unlimited.

Note that respawn only applies to automatic respawns and not the restart(8) command.

If the job has been respawned up to its respawn limit, the variable "$PROCESS" will be set to "respawn" to denote that the respawn limit was reached. See stopped(7) for further details.

6.30   script

Allows the specification of a multi-line block of shell code to be executed. Block is terminated by end script.

6.31   setgid

Added in Upstart v1.4


setgid <groupname>

Changes to the group <groupname> before running the job's process.


Note that all processes (pre-start, post-stop, et cetera) will be run with the group specified.

If this stanza is unspecified, the primary group of the user specified in the setuid block is used. If both stanzas are unspecified, the job will run with its group ID set to 0 in the case of system jobs, and as the primary group of the user in the case of User Jobs.


setgid apache

6.32   setuid

Added in Upstart v1.4


setuid <username>

Changes to the user <username> before running the job's process.


Note that all processes (pre-start, post-stop, et cetera) will be run as the user specified.

If this stanza is unspecified, the job will run as root in the case of system jobs, and as the user in the case of User Jobs.

Note that System jobs using the setuid stanza are still system jobs, and can not be controlled by an unprivileged user, even if the setuid stanza specifies that user.

Note that if you specify an invalid username in the setuid stanza, Upstart will log an error if it is in Debug Mode.

For example, if job foo specifies an invalid setuid username:

$ sudo initctl log-priority debug
$ sudo start foo
start: Job failed to start
$ sudo dmesg | grep setuid
[ 4942.908486] init: Failed to spawn foo main process: unable to find setuid user

Although the username is not logged, it is clear there is a problem with the setuid stanza for the specified foo job.

6.33   start on

This stanza defines the set of Events that will cause the Job to be automatically started.


start on EVENT [[KEY=]VALUE]... [and|or...]

Each event EVENT is given by its name. Multiple events are permitted using the operators "and" and "or" and complex expressions may be performed with parentheses (within which line breaks are permitted).

You may also match on the environment variables contained within the event by specifying the KEY and expected VALUE. If you know the order in which the variables are given to the event you may omit the KEY.

VALUE may contain wildcard matches and globs as permitted by fnmatch(3) and may expand the value of any variable defined with the env stanza.

Negation is permitted by using "!=" between the KEY and VALUE.

Note that if the job is already running and is not an instance job, if the start on condition becomes true (again), no further action will be taken.

Note that the start on stanza expects a token to follow on the same line. Thus:

# ERROR: invalid
start on
  foo or bar

# OK
start on foo or bar

If no environment variables are specified via KEY to restrict the match, the condition will match all instances of the specified event.

See Really understanding start on and stop on for further details.

6.33.1   Normal start

If you are just writing an upstart job that needs to start the service after the basic facilities are up, either of these will work:

start on (local-filesystems and net-device-up IFACE!=lo)


start on runlevel [2345]

The difference in whether to use the more generic 'runlevel' or the more explicit local-filesystems(7) and net-device-up events should be guided by your job's behaviour. If your service will come up without a valid network interface (for instance, it binds to, or uses setsockopt(2) SO_FREEBIND), then the runlevel event is preferable, as your service will start a bit earlier and start in parallel with other services.

However if your service requires that a non-loopback interface is configured for some reason (i.e., it will not start without broadcasting capabilities), then explicitly saying "once a non loopback device has come up" can help.

In addition, services may be aggregated around an abstract job, such as network-services:

start on started network-services

The network-services job is a generic job that most network services should follow in releases where it is available. [19] This allows the system administrator and/or the distribution maintainers to change the general startup of services that don't need any special case start on criteria.

We use the started(7) event so that anything that must be started before all network services can do "start on starting network-services".

6.33.2   Start depends on another service

start on started other-service

6.33.3   Start must precede another service

start on starting other-service

Example: your web app needs memcached to be started before apache:

start on starting apache2
stop on stopped apache2

exec /usr/sbin/memcached

6.34   stop on

This stanza defines the set of Events that will cause the Job to be automatically stopped if it is already running.


stop on EVENT [[KEY=]VALUE]... [and|or...]

Like the stop on stanza, start on expects a token to follow on the same line:

# ERROR: invalid
stop on
  foo or bar

# OK
stop on foo or bar

See start on for further syntax details.

6.34.1   Normal shutdown

stop on runlevel [016]

Or if a generic job is available such as network-services [19]

stop on stopping network-services

6.34.2   Stop before depended-upon service

stop on stopping other-service

Note that this also will stop when other-service is restarted, so you will generally want to couple this with the start on condition:

start on started other-service

6.34.3   Stop after dependent service

stop on stopped other-service

6.35   task

In concept, a task is just a short lived job. In practice, this is accomplished by changing how the transition from a goal of "stop" to "start" is handled.

Without the 'task' keyword, the events that cause the job to start will be unblocked as soon as the job is started. This means the job has emitted a starting(7) event, run its pre-start, begun its script/exec, and post-start, and emitted its started(7) event.

With task, the events that lead to this job starting will be blocked until the job has completely transitioned back to stopped. This means that the job has run up to the previously mentioned started(7) event, and has also completed its post-stop, and emitted its stopped(7) event.

Typically, task is for something that you just want to run and finish completely when a certain event happens.

# pre-warm-memcache

start on started memcached


exec /path/to/pre-warm-memcached

So you can have another job that starts your background queue worker once the local memcached is pre-warmed:

# queue-worker

start on stopped pre-warm-memcache
stop on stopping memcached


exec /usr/local/bin/queue-worker

The key concept demonstrated above is that we "start on stopped pre-warm-memcache". This means that we don't start until the task has completed. If we were to use started instead of stopped, we would start our queue worker as soon as /path/to/pre-warm-memcached had been started running.

We could also accomplish this without mentioning the pre-warm in the queue-worker job by doing this:

# queue-worker

start on started memcached
stop on stopping memcached


exec /usr/local/bin/queue-worker

# pre-warm-memcache

start on starting queue-worker
exec /path/to/pre-warm-memcache

If we did not use "task" in the above example, queue-worker would be allowed to start as soon as we executed /path/to/pre-warm-memcache, which means it might potentially start before the cache was warmed.

6.36   umask


umask <value>

Set the file mode creation mask for the process. <value> should be an octal value for the mask. See umask(2) for more details.


umask 0002

6.37   usage

Brief message explaining how to start the job in question. Most useful for instance jobs which require environment variable parameters to be specified before they can be started.


usage <string>


instance $DB
usage "DB - name of database instance"

If a job specifies the usage stanza, attempting to start the job without specifying the correct variables will display the usage statement. Additionally, the usage can be queried using initctl usage.

6.38   version


version <string>

This stanza may contain version information about the job, such as revision control or package version number. It is not used or interpreted by init(8) in any way.


version "1.0.2a-beta4"

7   Command-Line Options

The table below lists the command-line options accepted by the Upstart init daemon.


Under normal conditions, you should not need to specify any command-line options to Upstart. A number of these options were added specifically for testing Upstart itself and if used without due care can stop your system from booting (for example specifying --no-startup-event). Therefore you should be extremely careful specifying any command-line options to Upstart unless you understand the implications of doing so.

Command-line Options
Option Name Description Added in Version
--append-confdir=DIR Specify directory to read job configuration files from after the default(s). 1.13
--chroot-sessions Enable chroot sessions. 1.13
--confdir=DIR Specify alternate job configuration file directory (System Job default: /etc/init/) 1.3
--debug Enable Informational and debug messages 0.1.0
--default-console=VALUE Specify default value for jobs not specifying console (default: none (Upstart < 1.4), else log) 1.4
--help Show usage statement for init 0.1.0
--logdir=DIR Specify alternate log directory (System Job default: /var/log/upstart/) 1.4
--no-cgroups Make the cgroup stanza a NOP. 1.13
--no-dbus Stop PID 1 connecting to D-Bus system bus 1.11
--no-inherit-env Stop Session Jobs using the Session Init environment 1.7
--no-log Disable job logging (all job output is discarded) 1.4
--no-sessions Disable chroot sessions (name is historical). Removed in Upstart 1.13 (since the default is to disable). 1.3
--no-startup-event Disable emitting an event at startup 1.3
--prepend-confdir=DIR Specify directory to read job configuration files from before the default(s). 1.13
-q , --quiet Reduce output to errors only 0.1.0
--session Use D-Bus session bus rather than D-Bus system bus 1.3
--startup-event=NAME Specify an alternative initial event (default: startup event) 1.3
--user Run a Session Init 1.7
-v , --verbose Increase output to include informational messages 0.1.0
--version Display version information 0.1.0


8   Explanations

8.1   Really understanding start on and stop on

(Note: This section focuses on start on, but the information also applies to stop on unless explicitly specified).

The start on stanza needs careful contemplation. Consider this example:

start on started mysql

The syntax above is actually a short-hand way of writing:

start on started JOB=mysql

Remember that started(7) is an event which Upstart emits automatically when the mysql job has started to run. The whole start on stanza can be summarized as:

start on <event> [<vars_to_match_event_on>]

Where <vars_to_match_event_on> is optional, but if specified comprises one or more variables.

A slight variation of the above:

start on started JOB=mydb DBNAME=foobar

This example shows that the fictitious job above would only be started when the mydb database server brings the foobar database on-line. Correspondingly, file /etc/init/mydb.conf would need to specify "export DBNAME" and be started like this:

start mydb DBNAME=foobar

Looking at a slightly more complex real-life example:

# /etc/init/alsa-mixer-save.conf
start on starting rc RUNLEVEL=[06]

This job says,

"Run when the rc job emits the starting(7) event, but only if the
environment variable RUNLEVEL equals either 0 (halt) or 6 (reboot)".

If we again add in the implicit variable it becomes clearer:

# /etc/init/alsa-mixer-save.conf
start on starting JOB=rc RUNLEVEL=[06]

But where does the RUNLEVEL environment variable come from? Well, variables are exported in a job configuration file to related jobs. Thus, the answer is The rc Job.

If you look at this job configuration file, you will see, as deduced:


8.1.1   The rc Job

The rc job configuration file is well worth considering:

# /etc/init/rc.conf
start on runlevel [0123456]
stop on runlevel [!$RUNLEVEL]


console output


exec /etc/init.d/rc $RUNLEVEL

It says in essence,

"Run the SysV init script as /etc/init.d/rc $RUNLEVEL when telinit(8) emits the runlevel(7) event for any runlevel".

However, note the stop on condition:

stop on runlevel [!$RUNLEVEL]

This requires some explanation. The manual page for runlevel(7) explains that the runlevel event specifies two variables in the following order:


    The new "goal" runlevel the system is changing to.


    The previous system runlevel (which may be set to an empty value).

Thus, the stop on condition is saying:

"Stop the rc job when the runlevel event is emitted and the RUNLEVEL variable matches '[!$RUNLEVEL]'.

This admittedly does initially appear nonsensical. The way to read the statement above though is:

"Stop the rc job when the runlevel event is emitted and the RUNLEVEL variable is not set to the current value of the RUNLEVEL variable."

So, if the runlevel is currently "2" (full graphical multi-user under Ubuntu), the RUNLEVEL variable will be set to RUNLEVEL=2. The condition will thus evaluate to:

stop on runlevel [!2]

This is just a safety measure. What it is saying is:

  • if the rc job (which is a short-running Task) is still running when the system changes to a different runlevel (a runlevel other than "2" here), Upstart will stop it.
  • If it is not running when the system changes to a different runlevel, no action will be taken to stop the job (since it has already stopped).

However, note that when the system moves to a new runlevel, Upstart will then immediately re-run the job at the new runlevel since the start on condition specifies that this job should be started in every runlevel.

Since this job has specified the runlevel event, it automatically gets access to the variables set by this event (RUNLEVEL and PREVLEVEL). However, note that these two variables are also exported. The reason for this is to allow other jobs which start on or stop on the rc job to make use of these variables (which were set by the runlevel event).

See runlevel(7) for further details.

8.2   Environment Variables

Upstart allows you to set environment variables which will be accessible to the jobs whose job configuration files they are defined in. Environment variables are set using the env keyword.

For example:

# /etc/init/env.conf
env TESTING=123

  # prints "TESTING='123'" to system log
  logger -t $0 "TESTING='$TESTING'"
end script

Further, we can pass environment variables defined in events to jobs using the env stanza and the export stanza. Assume we have two job configuration files, A.conf and B.conf:

# /etc/init/A.conf
start on wibble
export foo

# /etc/init/B.conf
start on started A
  logger "value of foo is '$foo'"
end script

If we now run the following command, both jobs A and B will run, causing B to write "value of foo is 'bar'" to the system log:

# initctl emit wibble foo=bar

Note that a variables value can always be overridden by specifying a new value on the command-line. For example:

start on wibble
env var=hello

  logger "value of var is '$var'"
end script

When we emit the required event...:

# initctl emit wibble var=world

... the system log will have recorded:

value of var is 'world'

Note that a Job Configuration File does not have access to a user's environment variables, not even the superuser. This is not possible since all job processes created are children of init which does not have a user's environment.

However, using the technique above, it is possible to inject a variable from a user's environment into a job indirectly:

# initctl emit wibble foo=bar USER=$USER

As another example of environment variables, consider this job configuration file [20]:

env var=bar
export var

pre-start script
  logger "pre-start: before: var=$var"

  export var

  logger "pre-start: after: var=$var"
end script

post-start script
  logger "post-start: before: var=$var"

  export var

  logger "post-start: after: var=$var"
end script

  logger "script: before: var=$var"

  export var

  logger "script: after: var=$var"
end script

post-stop script
  logger "post-stop: before: var=$var"

  export var

  logger "post-stop: after: var=$var"
end script

This will generate output in your system log as follows (the timestamp and hostname have been removed, and the output formatted to make it clearer):

logger: pre-start:  before: var=bar
logger: pre-start:   after: var=pre-start

logger: post-start: before: var=bar
logger: post-start:  after: var=post-start

logger: script:     before: var=bar
logger: script:      after: var=main

logger: post-stop:  before: var=bar
logger: post-stop:   after: var=post-stop

As shown, every script section receives the value of $var as bar, but if any script section changes the value, it only affects that particular script sections copy of the variable. To summarize:

A script section cannot modify the value of a variable defined in a job configuration file for other script sections.

8.2.1   Restrictions

Environment variables do not expand in start on or stop on conditions:

env FOO=bar
start on $FOO

This will start the job in question when the "$FOO" event is emitted, not when the event "bar" is emitted:

# job above *NOT* started
initctl emit bar

# job above started!
initctl emit '$FOO'

Similarly, the following will not work:

start on starting $FOO
start on starting JOB=$FOO

8.2.2   Standard Environment Variables

The table below shows all variables set by Upstart itself. Note that variables prefixed by "UPSTART_" are variables set within a jobs environment, whereas the remainder are set within an events environment (see the following table).

Upstart Environment Variables.
Variable Brief Description Details
EXIT_SIGNAL Signal causing job to exit String such as "HUP" or "TERM", or numeric for unknown signals
EXIT_STATUS Exit code of job  
INSTANCE Instance name of $JOB Variable set but with no value if instance stanza not specified
JOB Name of job  
PROCESS Name of Job process type "main", "pre-start", "post-start", "pre-stop", "post-stop" or "respawn"
RESULT Whether job was successful "ok" or "failed"
UPSTART_EVENTS Events that caused job to start Space-separated. Event environment not provided
UPSTART_FDS File descriptor Number of the file descriptor corresponding to the listening socket-event(7) socket
UPSTART_JOB Name of current job  
UPSTART_SESSION Session Init D-Bus socket Allows initctl command to communicate with the appropriate Session Init
UPSTART_STOP_EVENTS Events that caused job to stop Space-separated. Event environment not provided

The following table lists the variables from the table above which are set when job events are emitted, and which are thus available from within a jobs environment.

Environment Variables by Event.
Event Variables Set in Event Environment
  • JOB
  • JOB
  • JOB
  • JOB

Notes that some variables (those marked with '*' and '') are only set when the job fails:

  • PROCESS will always be set.
  • Either EXIT_STATUS or EXIT_SIGNAL will be set.

Note carefully the distinction between JOB and UPSTART_JOB. If a job "bar.conf" specifies a start on condition of:

start on starting foo

and does not specify the instance stanza, when job "foo" starts, the environment of the "bar" job will contain:


8.3   Job with Multiple Duplicate Stanzas

The way in which Upstart parses the job configuration files means that "the last entry wins". That is to say, every job configuration file must be syntactically correct, but if you had a file such as:

start on event-A
start on starting job-B
start on event-C or starting job-D

This job will have a start on condition of:

start on event-C or starting job-D

...since that is the last start on condition specified.

For start on, stop on and emits stanzas, you can confirm Upstart's decision, you can use the initctl show-config command like this:

initctl show-config myjob

For the example above, the output would be:

start on event-C or starting job-D

9   Features

9.1   D-Bus Service Activation

As of D-Bus version 1.4.1-0ubuntu2 (in Ubuntu), you can have Upstart start a D-Bus service rather than D-Bus. This is useful because it is then possible to create Upstart jobs that start or stop when D-Bus services start.

See Run a Job When a User Logs in for an example.

10   Tools

Upstart provides a number of additional tools to:

10.1   Utilities

10.1.1   reload

Symbolically linked to initctl, causing the following to be run:

initctl reload <job>

This will send a running job the SIGHUP signal. By convention, daemons receiving this signal reload their configuration or in some way re-initialize themselves (keeping the same PID).

10.1.2   restart

Symbolically linked to initctl, causing the following to be run:

initctl restart <job>

Stops and then starts a job.

Note that restart is not the same as running stop followed by start since the restart command will retain the original job configuration whereas stopping the job and restarting it will load the latest job configuration from disk.

Further note that if the job contains post-stop, pre-start or post-start stanzas, these will NOT be run for a restart. However, a pre-stop stanza will be run.

10.1.4   start

Symbolically linked to initctl, causing the following to be run:

initctl start <job>

Starts a job.   Attempting to Start an Already Running Job

If you try to start a job that is already running and which does not specify the instance stanza, you will get the following error:

# start myjob
start: Job is already running: myjob   Attempting to Start a Job that requires an Instance Variable

If you try to start a job that specifies the instance stanza, you will need to specify the appropriate variable. If you do not, you will get an error. For example, assuming myjob.conf specified instance $foo:

# start myjob
start: Unknown parameter: foo

To resolve this, specify some value for the variable in question:

# start myjob foo="hello, world"

10.1.5   stop

Symbolically linked to initctl, causing the following to be run:

initctl stop <job>

Stops a job.   Attempting to Stop an Already Stopped Job

If you try to stop a job that is not running, you will get the following error:

# stop myjob
stop: unknown instance   Attempting to Stop a Job that requires an Instance Variable

If you try to stop a job that specifies the instance stanza without specifying the particular instance you wish to stop, you will get an error:

# stop myjob
stop: Unknown parameter: foo

To resolve this, specify the value for the variable in question:

# stop myjob foo=...

Where "..." must be replaced by a legitimate value for one of the instances as specified in the output of "initctl status myjob".

10.1.6   initctl

This is the primary command used by users and Administrators to interact with Upstart.

  • Run initctl help to see the available commands.
  • Run initctl --help to see the overall options available.
  • Run initctl <command> --help to see options for the specified command.

Commands to manipulate jobs:   initctl Commands Summary

Summary of initctl commands
Command Description Added in Version
initctl check-config Check for unreachable jobs/event conditions 1.3
initctl emit Emit an event 0.3.0
initctl get-env Retrieve a variable from the job environment table 1.7
initctl help Display list of commands 0.3.0
initctl list List known jobs 0.2.0
initctl list-env List job environment table 1.7
initctl list-sessions List running User Sessions 1.7
initctl log-priority Change the minimum priority of log messages displayed by the init daemon 0.3.8
initctl notify-cgroup-manager-address Inform Upstart of address of cgroup manager 1.13
initctl notify-disk-writeable Inform Upstart that disk is now writeable 1.5
initctl reload Send HUP signal to job 0.6.5
initctl reload-configuration Reload the configuration 0.6.0
initctl restart Restart job 0.6.0
initctl reset-env Revert the the job environment table to its default values 1.7
initctl set-env Store a variable from the job environment table 1.7
initctl show-config Show emits, start on and stop on details for job(s) 1.3
initctl start Start job 0.1.0
initctl status Query status of job 0.1.0
initctl stop Stop job 0.1.0
initctl unset-env Remove a variable from the job environment table 1.7
initctl usage Show job usage message if available 1.5
initctl version Request the version of the init daemon 0.3.8   initctl check-config

The initctl check-config command can be used to check that the events and jobs a job configuration file references are "known" to the system. This is important, since if a System Administrator were to inadvertently force the removal of a package, or inadvertently delete a critical job configuration file, the system may no longer boot. Usage is simple:

$ # search all job configuration files for "unreachable" conditions
$ initctl check-config

$ # search specified job configuration file for unreachable conditions
$ initctl check-config <job>

Some job configuration files -- such as plymouth.conf -- have complex start on conditions which look for any of a number of jobs. As long as one valid set of events can be satisfied, check-config will be happy. However, to see if it found any missing jobs or events, specify the --warn option. Note that the first invocation returns no output, denoting that no problems have been found:

$ initctl check-config plymouth
$ initctl check-config --warn plymouth
  start on: unknown job uxlaunch
  start on: unknown job lightdm
  start on: unknown job lxdm
  start on: unknown job xdm
  start on: unknown job kdm

Note that this is not an error condition since although check-config cannot satisfy any of these jobs, it can satisfy the overall configuration for plymouth (by the gdm job - see plymouth.conf on Ubuntu).

Note that the check-config command relies on the emits stanza to be correctly specified for each job configuration file that emits an event (see init(5)). See also [30].   initctl emit

Generates an arbitrary event.


# initctl emit hello-world


If you attempt to emit an event and it blocks (appears to hang), this is because there are other jobs which have a start on or stop on condition which contains this event. See Event Types for further details.   initctl get-env

Retrieve value of a job environment variable. See Job Environment.


$ initctl get-env foo
$   initctl help

Displays a list of initctl commands.   initctl list

The list command simply aggregates the status of all job instances. See initctl status.


# show all Session Jobs
$ initctl list

# show all System Jobs
$ initctl --system list   initctl list-env

Show all variables in the job environment table. See Job Environment.

Note that all Session Jobs inherit the environment of the Session Init.


  • In Upstart 1.7 and 1.8 the job environment table only contained variables explicitly set by initctl set-env.
  • In Upstart 1.9 and above, the job environment table contains all variables inherited by the Session Init.   initctl list-sessions

List all running sessions being managed by a Session Init. The format of this command is:

<pid> <socket>

Where <pid> is the process ID of the running Session Init instance and <socket> is the private D-Bus socket address the Session Init listens on.   initctl log-priority

To change the priority with which Upstart logs messages to the system log, you can change the log priority at any time using log-priority command as follows:

initctl log-priority <priority>

Where <priority> may be one of:

  • debug
  • info
  • message
  • warn
  • error
  • fatal

For example:

# same as "--verbose"
$ sudo initctl log-priority info

# same as "--debug"
$ sudo initctl log-priority debug

The default priority is message:

$ initctl log-priority

If the log-priority is changed, it can be reverted to the default like this:

# return to default value
$ sudo initctl log-priority message

Note that you will need to check the configuration for your system logging daemon (generally syslog(3) or rsyslogd(8)) to establish where it logs the output.

the output of these options is handled by your systems look at the particular daemons configuration to know where to find the output.

For a standard Ubuntu Maverick (10.10) system, the output will be sent to file /var/log/daemon.log, whilst on newer Ubuntu systems such as Ubuntu Natty (11.04), the output will be directed to file /var/log/syslog.   initctl notify-cgroup-manager-address

Command to inform Upstart of the address the cgroup manager is listening on. See cgmanager(8).   initctl notify-disk-writeable

Command that is used to notify Upstart that the log disk is writeable [11].

This is an indication to Upstart that it can flush the log of job output for jobs that ended before the log disk became writeable. If logging is enabled, this command must be called once the disks become writeable.   initctl reload

Causes the SIGHUP signal to be sent to the main job process since this signal is commonly used to inform an application to re-initialize itself. Note that the jobs associated Job Configuration File is not re-read.   initctl reload-configuration

Force the init daemon to reload its configuration files.

It is generally not necessary to call this command since the init daemon watches its configuration directories with inotify(7) and automatically reloads in cases of changes.

Note that no jobs will be started by this command.   initctl reset-env

Applies to Session Jobs only.

Return the job environment table to its default values. See Job Environment.   initctl restart

Cause the associated job to be killed and respawned. Note that this does not cause the job to re-read its Job Configuration File: to force this, stop the job and then start it.   initctl set-env

Applies to Session Jobs only.

Adds or updates a variable in the job environment table. See Job Environment.

Note that as of upstart 1.13, multiple name/value pairs may be specified.


$ initctl set-env foo='hello world'   initctl show-config

The initctl show-config command can be used to display details of how Upstart has parsed one or more job configuration files. The command displays the start on, stop on and emits stanzas. This might seem rather pointless, but it is extremely useful since:

  • The command will fully-bracket all start on and stop on conditions.

    This shows how Upstart has parsed complex conditions. For example, if job myjob specified a start on condition:

    start on starting a or b and stopping c or d

    The command would return:

      start on (((starting a or b) and stopping c) or d)
  • The command can produce machine parseable output showing the types of entities by specifying the "--enumerate" option.

    For example, the job above would be displayed as:

      start on starting (job: a, env:)
      start on b (job:, env:)
      start on stopping (job: c, env:)
      start on d (job:, env:)


    • a is a job (with triggering event starting(7)).
    • b is an event.
    • c is a job (with triggering event stopping(7)).
    • d is an event.
  • The command shows the environment for the events.

    Assuming a (ridiculous) start on condition of:

    start on event-a foo=bar a=b c=22 d="hello world" or stopped job-a e=123 f=blah or hello world=2a or starting foo foo=foo


    $ initctl show-config --enumerate myjob
      start on event-a (job:, env: foo=bar a=b c=22 d=hello world)
      start on stopped (job: job-a, env: e=123 f=blah)
      start on hello (job:, env: world=2a)
      start on starting (job: foo, env: foo=foo)

    As shown, this makes the condition (slightly!) easier to understand:

    • event-a is an event with 4 environment variables:
      • foo=bar
      • a=b
      • c=22
      • d=hello world
    • job-a is a job with triggering event stopped(7) and 2 environment variables:
      • e=123
      • f=blah
    • hello is an event with 1 environment variable:
      • world=2a
    • foo is a job with triggering event starting(7) and 1 environment variable:
      • foo=foo

See also [29].   initctl start

Start the specified job or job instance.   initctl status

The status(8) command shows the status of all running instances of a particular job.

Prior to Upstart 1.7, running this command as a non-privileged user would list both System Jobs and User Jobs (see System Job and User Job). However, as of Upstart 1.7, if this command (or initctl list) is run from within a session, it will only list Session Jobs.

To see system jobs from within a session, specify the "--system" command-line option like this:

# show Session Jobs
$ initctl status myjob

# show System Jobs
$ initctl --system somejob

The format of the output can be summarized as follows:

<job> [ (<instance>)]<goal>/<status>[, process <PID>]
        [<section> process <PID>]

Considering each field:

  • <job> is the name of the job

    Essentially, this is the name of the job configuration file, less the path and without the ".conf" extension. Thus, /etc/init/myjob.conf would display as "myjob".

  • <instance> is the job instance.

    See instance and Determining How to Stop a Job with Multiple Running Instances.

  • <goal>

    Every job has a goal of either start or stop where the goal is the target the job is aiming for. It may not achieve this target, but the goal shows the "direction" the job is heading in: it is either trying to be started, or be stopped.

    • When a Task Job starts, its goal will be start and once the task in question has completed, Upstart will change its goal to stop.
    • When a Service Job starts, its goal will be start and will remain so until either the jobs stop on condition becomes true, or an Administrator manually stops the job using stop.
  • <status>

    The job instances status. See Job States.

  • <PID> is the process ID of the running process corresponding to <job>.

    See ps(1).

  • <section> is a script or exec section (such as pre-stop).

Lets look at some examples...   Single Job Instance Running without PID

Here is the summarised syntax:

<job> <goal>/<status>


ufw start/running

You may be forgiven for thinking this rather curious specimen is an Abstract Job. Although you cannot determine the fact from the output above, this job is not an abstract job. If you look at its job configuration file /etc/init/ufw.conf, you'll see the following:

description     "Uncomplicated firewall"

# Make sure we start before an interface receives traffic
start on (starting network-interface
          or starting network-manager
          or starting networking)

stop on runlevel [!023456]

console output

pre-start exec /lib/ufw/ufw-init start quiet
post-stop exec /lib/ufw/ufw-init stop

Notice the last two lines above. The firewall job configuration file has a pre-start section and a post-stop section, but no script or exec section. So, once Upstart has run the pre-start command and the job is "running", it won't actually have a PID (since the pre-start command will have finished and there is no further command to run until the job stops).   Single Job Instance Running Job with PID

A single instance of a running job can be summarized like this:

<job> <goal>/<status>, process <PID>

This is possibly the "most common case" of jobs you will see. For example:

cups start/running, process 1733


  • <job> is "cups" (/etc/init/cups.conf).
  • <goal> is "start"
  • <status> is "running"
  • <process> is "1733" (as shown by ps(1)).   Single Job Instance Running with Multiple PIDs

This can be summarized as:

<job> <goal>/<status>, process <PID>
  <section> process <PID>

For example:

ureadahead stop/pre-stop, process 227
      pre-stop process 5579

What is going on here? Picking this apart we have:

  • ureadahead is the job (/etc/init/ureadahead.conf).
  • stop is the goal (job is trying to stop).
  • pre-stop is the job status (it is running the pre-stop section as PID 5579).
  • the script or exec stanza is also running under PID 227. See pre-stop for further details.   Multiple Running Job Instances Without PID


<job> (<instance>) <goal>/<status> (<instance>)
<job> (<instance>) <goal>/<status> (<instance>)

A job with multiple instances might look a little strange initially. Here is an example:

network-interface (lo) start/running
network-interface (eth0) start/running


  • network-interface is the job (/etc/init/network-interface.conf).
  • job instances are:
    • lo
    • eth0
  • start is the goal (job instances are currently running).
  • running is the job status (it is running).

A slightly more complex example:

network-interface-security (network-manager) start/running
network-interface-security (network-interface/eth0) start/running
network-interface-security (network-interface/lo) start/running
network-interface-security (networking) start/running


  • network-interface-security is the job (/etc/init/network-interface-security.conf).
  • job instances are:
    • network-manager
    • network-interface/eth0
    • network-interface/lo
    • networking
  • start is the goal (job instances are currently running).
  • running is the job status (it is running).

Let's look at the main elements of the corresponding job configuration file:

start on (starting network-interface
          or starting network-manager
          or starting networking)

instance $JOB${INTERFACE:+/}${INTERFACE:-}

pre-start script
  # ...
end script

Again, this job has no script or exec section, but it does have a pre-start script section. Also, note the interesting instance stanza. This explains the rather odd-looking instance names listed above.   Multiple Running Job Instances With PIDs


<job> (<instance>) <goal>/<status> (<instance>), process <PID>

For example:

foo (1) start/running, process 30003
foo (hello 1,2,3) start/running, process 30008


  • foo is the job (/etc/init/foo.conf).
  • start is the goal (it is not trying to stop).
  • running is the job status (it is running).
  • instances are:
    • 1 (PID 30003)
    • hello 1,2,3 (PID 30008)   Multiple Running Job Instances With Multiple PIDs


<job> (<instance>) <goal>/<status> (<instance>), process <PID>
        <section> process <PID>

For example:

myjob (foo) stop/pre-stop, process 31677
        pre-stop process 31684
myjob (bar) stop/pre-stop, process 31679
        pre-stop process 31687
myjob (bzr) stop/pre-stop, process 31681
        pre-stop process 31690


  • myjob is the job (/etc/init/myjob.conf).
  • stop is the goal (job is trying to stop).
  • pre-stop is the job status (it is running the pre-stop section for each instance).
  • instances are:
    • foo (PID 31677, with pre-stop PID 31684)
    • bar (PID 31679, with pre-stop PID 31687)
    • baz (PID 31681, with pre-stop PID 31690)

It is instructive to see how we got to the output above. Here is the job configuration file:

instance $foo

exec sleep 999

pre-stop script
  sleep 999
end script

We then started three instances like this:

# for i in foo bar baz; do start -n myjob foo=$i; done

Note we used the "-n" option to start to ensure we didn't have to wait for each instance to complete before starting the next.

Now all three instances are running:

# initctl list|grep -A 1 ^inst
myjob start/running (foo), process 31677
myjob start/running (bar), process 31679
myjob start/running (baz), process 31681

To trigger the pre-stop, we need to stop the instances:

# for i in foo bar baz; do stop -n myjob foo=$i; done
myjob (foo) stop/pre-stop, process 31677
      pre-stop process 31684
myjob (bar) stop/pre-stop, process 31679
      pre-stop process 31687
myjob (baz) stop/pre-stop, process 31681
      pre-stop process 31690

Now, running initctl will show the output at the start of this section.   Stopped Job


<job> <goal>/<status>

A job that is not running (has no instances):

rc stop/waiting


  • rc is the job (/etc/init/rc.conf).
  • stop is the goal (it is not trying to start).
  • waiting is the job status (it is not running).   initctl stop

Stop the specified job or job instance.   initctl unset-env

Applies to Session Jobs only.

Discards the specified variable from the job environment table. See Job Environment.

Note that as of Upstart 1.9, any variable inherited from when the Session Init starts can be unset.

Note that as of Upstart 1.13, multiple name/value pairs may be specified.   initctl usage

This command allows the usage for a job to be queried:

$ initctl usage <job>

Note that if a job is specified which does not use the usage stanza, no usage will be displayed.   initctl version

Display the version of the init daemon. To display the version of initctl itself, run:

initctl --version

10.1.7   init-checkconf

The init-checkconf script performs checks on a job configuration file prior to installing it in /etc/init/. The script must be run as a non-root user for all versions prior to that provided by Upstart 1.12.

To ensure that you haven't misused the Upstart syntax, use the init-checkconf command:

$ init-checkconf myjob.conf

See init-checkconf(8) for further details.

10.1.8   upstart-monitor

Added in Upstart v1.8 (requires at least Upstart v1.7).

The upstart-monitor is a utility that is used to display Upstart events as they are emitted.

It is useful to:

The utility can run either as a command-line (CLI) application:

upstart-monitor in CLI mode.

It can also run as as a GUI:

upstart-monitor in GUI mode.

If the required GUI libraries are not available, it will automatically run in CLI mode.

10.1.9   mountall ( , )

NOTE: mountall(8) is a Debian and Ubuntu specific extension.

The mountall daemon is the program that mounts your filesystems during boot on an Ubuntu system. It does this by parsing both /etc/fstab and its own fstab file /lib/init/fstab, and mounting the filesystems it finds listed. Additionally, it handles running fsck(8).

See fstab(5).   Mountall events

Mountall also emits a number of useful events. For every filesystem it determines needs to be mounted, it will emit up to 2 events:

  • mounting
  • mounted

Additional to the couplet above, mountall also emits the following "well-known" events. The sections below provide details.

The mountall daemon is unusual in emitting such a number of events. However, it does this to provide as much flexibility as possible since making disks and filesystem available is such an important part of the boot process (and a lot of other jobs need to be notified when certain mounts become available).   mounting

Emitted when a particular filesystem is about to be mounted.

See mounting(7).   mounted

Emitted by when a particular filesystem has been mounted successfully.

Note that if a filesystem failed to mount, no corresponding mounted event will be emitted.

See mounted(7).   all-swaps

Emitted when all swap devices are mounted.

See all-swaps(7).   filesystem

Emitted after mountall (debian-and-ubuntu-specific) has mounted (or at least attempted to mount) all filesystems.

See filesystem(7).   virtual-filesystems

Emitted after the last virtual filesystem has been mounted.

See virtual-filesystems(7).   local-filesystems

Emitted after the last local filesystem has been mounted.

See local-filesystems(7).   remote-filesystems

Emitted after the last remote filesystem has been mounted.

See remote-filesystems(7).   Mountall Event Summary

|mounting MOUNTPOINT=/virtual-1                              |  mounting TYPE=swap |
|mounted  MOUNTPOINT=/virtual-1                              |  mounted  TYPE=swap |
|     :                                                      |  all-swaps          |
|mounting MOUNTPOINT=/virtual-n                              |                     |
|mounted  MOUNTPOINT=/virtual-n                              |                     |
|virtual-filesystems                                         |                     |
+-----------------------------+------------------------------+                     |
|mounting MOUNTPOINT=/local-1 |mounting MOUNTPOINT=/remote-1 |                     |
|mounted  MOUNTPOINT=/local-1 |mounted  MOUNTPOINT=/remote-1 |                     |
|     :                       |     :                        |                     |
|mounting MOUNTPOINT=/local-n |mounting MOUNTPOINT=/remote-n |                     |
|mounted  MOUNTPOINT=/local-n |mounted  MOUNTPOINT=/remote-n |                     |
|local-filesystems            |remote-filesystems            |                     |
|filesystem                                                                        |

The diagram above shows the different event flows when mountall runs. Note in particular that columns should be considered as independent "threads" of execution (can happen at any time and independently), and rows are sequential: rows lower down the chart occur at at later time than those higher up the chart.

Notes on mountall event emission:

  • swap partitions are processed at any time.
  • virtual filesystems are processed at any time.
  • virtual filesystems are processed before local or remote filesystems (regardless of their ordering in /etc/fstab).
  • local and remote filesystems are mounted at any time after the last virtual filesystem has been mounted.

See mounting(7) and mounted(7). For a concise summary of all available events generated by mountall, see upstart-events(7).   mountall Examples

The examples which follow were generated using the following job configuration file /etc/init/get_mountall.conf:

start on (local-filesystems
      or (mounting
      or (mounted
      or (virtual-filesystems
      or (remote-filesystems
      or (all-swaps or filesystem))))))

  echo "\n`env`" >> /dev/.initramfs/mountall.log
end script

Script output:

























10.2   Bridges

Bridges react to events from some other (non-Upstart) source and create corresponding Upstart events.

Summary of available bridges.
Bridge Added in Version
upstart-dbus-bridge 1.9
upstart-dconf-bridge 1.10
upstart-event-bridge 1.7
upstart-file-bridge 1.8
upstart-local-bridge 1.10
upstart-socket-bridge 1.3 (bundled for first time)
upstart-udev-bridge 1.3 (bundled for first time)

10.2.1   plymouth-upstart-bridge ()

The plymouth-upstart-bridge is an Ubuntu-specific facility to allow Plymouth to display Upstart state changes on the boot splash screen.

See the Plymouth Ubuntu wiki page for more information on Plymouth.

10.2.2   upstart-socket-bridge

The Upstart socket bridge is an out-of-process application that "listens" for jobs that announce they "start on socket". The bridge arranges for the jobs in question to be started automatically at the point the first client connection is made on the socket specified in their start on condition. See socket-event(7).

This is a useful "lazy" facility in that it allows for applications which are expensive to load to be started "on demand" rather than simply at some point on every boot: if you have no customers to your web site one day, there is probably no point in starting your database server. The downside to using the bridge being that the first client connection will probably be slower than subsequent connections to allow the application time to start.

Supported socket types:

  • Unix sockets.
  • Abstract sockets.
  • IPv4 sockets.
  • IPv6 sockets (requires version provided with Upstart 1.12, or above).

10.2.3   upstart-udev-bridge

The Upstart udev(7) bridge creates Upstart events from udev events. As documented in upstart-udev-bridge(8), Upstart will create events named:



  • <subsystem> is the udev subsystem.
  • <action> is the udev action.

Upstart maps the three actions below to new names, but any other actions are left unmolested:

  • add becomes added
  • change becomes changed
  • deleted becomes removed

To see a list of possible Upstart events for your system:

for subsystem in /sys/class/*
  for action in added changed removed
    echo "${subsystem}-device-${action}"

Alternatively, you could parse the following:

# udevadm info --export-db

To monitor udev events:

$ udevadm monitor --environment

And now for some examples...

If a job job-A specified a start on condition of:

start on (graphics-device-added or drm-device-added)

To see what sort of information is available to this job, we can add the usual debugging information:

start on (graphics-device-added or drm-device-added)
  echo "`env`" > /dev/.initramfs/job-A.log
end script

Here is an example of the log:


Another example specifying a start on containing net-device-added:

ID_VENDOR_FROM_DATABASE=Realtek Semiconductor Co., Ltd.

Plugging in a USB webcam will generate an input-device-added event:

DEVLINKS=/dev/char/13:76 /dev/input/by-path/pci-0000:00:1d.0-event

Note: you may get additional events if it also includes a microphone or other sensors.

Plugging in a USB headset (headphones plus a microphone) will probably generate three events:

  • sound-device-added (for the headphones):

  • usb-device-added (also for the headphones):

    UPOWER_VENDOR=Logitech, Inc.
  • input-device-added (for the microphone):

    NAME="Logitech Logitech USB Headset"
    KEY==c0000 0 0 0
    PWD=/   Careful Use of udev Events

You need to be careful when using the upstart-udev-bridge since certain devices are NOT ready at the point the kernel generates the original udev event: in these circumstances, all the kernel is saying is "I have this device", not "I have this device and it is ready to use".

The problem is that the kernel does not know when the device is ready and neither can Upstart know this. The kernel is simply signalling that the device has either:

  • become available (once the upstart-udev-bridge emits the "*-device-added" event).
  • changed state somehow (once the upstart-udev-bridge emits the one or more "*-device-changed" events).

So, for example, just because you have received a "usb-device-added" event for your USB modem does not guarantee that the modem is operational.

Unfortunately, every device acts differently, so you really do need specialist knowledge of the device in question.

However, a general rule of thumb is that a device is ready once Upstart has emitted a "changed" event for the device which also includes a "ID_" variable in that events environment. This is of particular importance for "block" devices and "sound" devices.

10.2.4   upstart-event-bridge

Added in Upstart v1.7

An instance of the upstart-event-bridge runs for each logged in user and proxies system-level events down to the users session. In plain English, this means all events emitted by a system job become visible to jobs running as the user.

To allow Session Jobs to distinguish between User Events and System Events, the upstart-event-bridge prefixes all system events with ":sys:". So for example, when the "foo" system job starts, at the system level the following event will be emitted:

started JOB=foo

That event is visible to all System Jobs, but is invisible to Session Jobs. However, the following event will be emitted by the upstart-event-bridge to allow Session Jobs to react to the "foo" system job starting:

:sys:started JOB=foo

10.2.5   upstart-file-bridge

Added in Upstart v1.8

The upstart-file-bridge allows jobs to react to file events. It currently uses inotify(7) and is available both for System Jobs and Session Jobs.


start on file FILE=PATH EVENT=TYPE [MATCH=PATH]   Examples

Start a job when file is created, modified or deleted:

start on file FILE=/run/

Start job when file is created (only):

start on file FILE=/run/ EVENT=create

Start job when any files within a directory are created, modified or deleted:

start on file FILE=/var/log/

Start job when files that match a glob pattern are created in the indicated directory:

start on file FILE=/var/crash/*.crash EVENT=create

For more details, see upstart-file-bridge(8) and file-event(7).

10.2.6   upstart-dbus-bridge

Added in Upstart v1.9

The upstart-dbus-bridge allows jobs to react to D-Bus signals.



Start a job when D-Bus signal NameAcquired is received:

start on dbus SIGNAL=NameAcquired INTERFACE=org.freedesktop.DBus PATH=/org/freedesktop/DBus SENDER=org.freedesktop.DBus

See upstart-dbus-bridge(8) and dbus-event(7) for further details.

10.2.7   upstart-dconf-bridge

Added in Upstart v1.10

The upstart-dconf-bridge is a Session Init only bridge that allows Session Jobs to react to DConf database changes.

10.2.8   upstart-local-bridge

Added in Upstart v1.10

The upstart-local-bridge is a specialist bridge that allows jobs to react to name/value pairs sent to a local socket created by the bridge.

11   Cookbook and Best Practises

11.1   List All Jobs

To list all jobs on the system along with their states, run:

$ initctl list

See initctl.

11.2   List All Jobs With No stop on Condition

# list all jobs (stopped and running instances), and compact down
# to actual job names.
initctl list | awk '{print $1}' | sort -u | while read job
  # identify jobs with no "stop on"
  initctl show-config -e $job | grep -q "^  stop on" || echo "$job"

11.3   List All Events That Jobs Are Interested In On Your System

Here is another example of how initctl show-config can be useful:

initctl show-config -e | egrep -i "(start|stop) on" | awk '{print $3}' | sort -u

11.4   Create an Event

To create, or "emit" an event, use initctl(8) specifying the emit command.

For example, to emit the hello event, you would run:

# initctl emit hello

This event will be "broadcast" to all Upstart jobs.

If you are creating a job configuration file for a new application, you probably do not need to do this though, since Upstart emits events on behalf of a job whenever the job changes state.

A simple configuration file like that shown below may suffice for your application:

# /etc/init/myapp.conf
description "run my app under Upstart"
exec /path/to/myapp

11.5   Create an Event Alias

Say you have an event, but want to create a different name for it, you can simulate a new name by creating a new job which:

  • has a start on that matches the event you want to "rename"
  • is a task
  • emits the new name for the event

For example, if you wanted to create an alias for a particular flavour of the runlevel event called "shutdown" which would be emitted when the system was shutdown, you could create a job configuration file called /etc/init/shutdown.conf containing:

start on runlevel RUNLEVEL=0
exec initctl emit shutdown

Note that this isn't a true alias since:

  • there are now two events which will be generated when the system is shutting down:
    • runlevel RUNLEVEL=0
    • shutdown
  • the two events will be delivered by Upstart at slightly different times (shutdown will be emitted just fractionally before runlevel RUNLEVEL=0).

However, the overall result might suffice for your purposes such that you could create a job configuration file like the following which will run (and complete) just before your system changes to runlevel 0 (in other words halts):

start on shutdown

11.5.1   Change the Type of an Event

Note that along with creating a new name for an event, you could make your alias be a different type of event. See Event Types for further details.

11.6   Synchronisation

Upstart is very careful to ensure when a condition becomes true that it starts all relevant jobs in sequence (see Order in Which Jobs Which start on the Same Event are Run). However, although Upstart has started them one after another they might still be running at the same time. For example, assume the following:

  • /etc/init/X.conf

    start on event-A
      echo "`date`: $UPSTART_JOB started" >> /tmp/test.log
      sleep 2
      echo "`date`: $UPSTART_JOB stopped" >> /tmp/test.log
    end script
  • /etc/init/Y.conf

    start on event-A
      echo "`date`: $UPSTART_JOB started" >> /tmp/test.log
      sleep 2
      echo "`date`: $UPSTART_JOB stopped" >> /tmp/test.log
    end script
  • /etc/init/Z.conf

    start on event-A
      echo "`date`: $UPSTART_JOB started" >> /tmp/test.log
      sleep 2
      echo "`date`: $UPSTART_JOB stopped" >> /tmp/test.log
    end script

Running the following will cause all the jobs above to run in some order:

# initctl emit event-A

Here is sample output of /tmp/test.log:

Thu Mar 31 10:20:44 BST 2011: Y started
Thu Mar 31 10:20:44 BST 2011: X started
Thu Mar 31 10:20:44 BST 2011: Z started
Thu Mar 31 10:20:46 BST 2011: Y stopped
Thu Mar 31 10:20:46 BST 2011: Z stopped
Thu Mar 31 10:20:46 BST 2011: X stopped

There are a few points to note about this output:

  • All jobs start "around the same time" but are started sequentially.
  • The order the jobs are initiated by Upstart cannot be predicted.
  • All three jobs are running concurrently.

It is possible with a bit of thought to create a simple framework for synchronisation. Take the following job configuration file /etc/init/synchronise.conf:


This one-line Abstract Job configuration file is extremely interesting in that:

  • Since it includes the manual keyword, a job created from it can only be started manually.
  • Only a single instance of a job created from this configuration can exist (since no instance stanza has been specified).

What this means is that we can use a job based on this configuration as a simple synchronisation device.

The astute reader may observe that synchronise has similar semantics to a POSIX pthread condition variable.

Now we have our synchronisation primitive, how do we use it? Here is an example which we'll call /etc/init/test_synchronise.conf:

start on stopped synchronise

# allow multiple instances
instance $N

# this is not a service

pre-start script
  # "lock"
  start synchronise || true
end script

  # do something here, knowing that you have exclusive access
  # to some resource that you are using the "synchronise"
  # job to protect.
  echo "`date`: $UPSTART_JOB ($N) started" >> /tmp/test.log
  sleep 2
  echo "`date`: $UPSTART_JOB ($N) stopped" >> /tmp/test.log
end script

post-stop script
  # "unlock"
  stop synchronise || true
end script

For example, to run 3 instances of this job, run:

for n in $(seq 3)
  start test_synchronise N=$n

Here is sample output of /tmp/test.log:

Thu Mar 31 10:32:20 BST 2011: test_synchronise (1) started
Thu Mar 31 10:32:22 BST 2011: test_synchronise (1) stopped
Thu Mar 31 10:32:22 BST 2011: test_synchronise (2) started
Thu Mar 31 10:32:24 BST 2011: test_synchronise (2) stopped
Thu Mar 31 10:32:25 BST 2011: test_synchronise (3) started
Thu Mar 31 10:32:27 BST 2011: test_synchronise (3) stopped

The main observation here:

  • Each instance of the job started and stopped before any other instance ran.

Like condition variables, this technique require collaboration from all parties. Note that you cannot know the order in which each instance of the test_synchronise job will run.

Note too that it is not necessary to use instances here. All that is required is that your chosen set of jobs all collaborate in their handling of the "lock". Instances make this simple since you can spawn any number of jobs from a single "template" job configuration file.

11.7   Determine if Job was Started by an Event or by "start"

A job that specifies a start on condition can be started in two ways:

  • by Upstart itself when the start on condition becomes true.
  • by running, "start <job>".

Interestingly, it is possible for a job to establish how it was started by considering the UPSTART_EVENTS variable:

  • If the UPSTART_EVENTS variable is set in the job environment, the job was started by an event.
  • If the UPSTART_EVENTS variable is not set in the job environment, the job was started by the start command.

Note that this technique does not allow you to determine definitively if the job was started manually by an Administrator since it is possible that if the UPSTART_EVENTS variable is not set that the job was started by another job calling start inside a script section.

11.8   Stop a Job from Running if A pre-start Condition Fails

If you wish a job to not be run if a pre-start condition fails:

pre-start script
  # main process will not be run if /some/file does not exist
  test -f /some/file || { stop ; exit 0; }
end script

  # main process is run here
end script

11.9   Run a Job Only When an Event Variable Matches Some Value

By default, Upstart will run your job if the start on condition matches the events listed:

start on event-A

But if event-A provides a number of environment variables, you can restrict your job to starting only when one or more of these variables matches some value. For example:

start on event-A FOO=hello BAR=wibble

Now, Upstart will only run your job if all of the following are true:

  • the event-A is emitted
  • the value of the $FOO variable in event-A's environment is "hello".
  • the value of the $BAR variable in event-A's environment is "wibble".

11.10   Run a Job when an Event Variable Does Not Match Some Value

Upstart supports negation of environment variable values such that you can say:

start on event-A FOO=hello BAR!=wibble

Now, Upstart will only run your job if all of the following are true:

  • the event-A is emitted
  • the value of the $FOO variable in event-A's environment is "hello".
  • the value of the $BAR variable in event-A's environment is not "wibble".

11.11   Run a Job as Soon as Possible After Boot

(Note: we ignore the initramfs in this section).

To start a job as early as possible, simply "start on" the startup event. This is the first event Upstart emits and all other events and jobs follow from this:

start on startup

11.12   Run a Job When a User Logs in Graphically ()

Assuming a graphical login, this can be achieved using a start on condition of:

start on desktop-session-start

This requires the display manager emit the event in question. See the upstart-events(7) man page on an Ubuntu system for the 2 events a Display Manager is expected to emit. If your Display Manager does not emit these event, check its documentation to see if it allows scripts to be called at appropriate points and then you can easily conform to the reference implementations behaviour:

# A user has logged in
/sbin/initctl -q emit desktop-session-start \

# Display Manager has initialized and displayed a login screen
# (if appropriate)
/sbin/initctl -q emit login-session-start \

11.13   Run a Job When a User Logs in

This makes use of D-Bus Service Activation.

  1. Add "UpstartJob=true" to file "/usr/share/dbus-1/system-services/org.freedesktop.ConsoleKit.service".

  2. Create a job configuration file corresponding to the D-Bus service, say /etc/init/user-login.conf [16]:

    start on dbus-activation org.freedesktop.ConsoleKit
    exec /usr/sbin/console-kit-daemon --no-daemon
  3. Ensure that the D-Bus daemon ("dbus-daemon") is started with the --activation=upstart option (see /etc/init/dbus.conf).

Now, when a user logs in, D-Bus will emit the dbus-activation event, specifying the D-Bus service started. You can now create other jobs that start on user-login.

11.13.1   Environment

Below is an example of the environment such an Upstart D-Bus job runs in:


11.14   Run a Job For All of a Number of Conditions

If you have a job configuration file like this:

start on (event-A or (event-B or event-C))

  echo "`date`: ran in environment: `env`" >> /tmp/myjob.log
end script

Upstart will run this job when any of the following events is emitted:

  • event-A
  • event-B
  • event-C

You cannot know the order in which the events will arrive in, but the specified start on condition has told Upstart that any of them will suffice for your purposes. So, if event-B is emitted first, Upstart will run the job and only consider re-running the job if and when the job has finished running. If event-B is emitted and the job is running and then (before the job finishes running) event-A is emitted, the job will not be re-run.

However, what if you wanted to run the script for all the events? If you know that all of these events will be emitted at some point, you could change the start on to be:

start on (event-A and (event-B and event-C))

Here, the job will only run at the time when the last of the three events is received.

Is it possible to run this job for each event as soon as each event arrives? Yes it is:

start on (event-A or (event-B or event-C))


  echo "`date`: ran in environment: `env`" >> /tmp/myjob.log
end script

By adding the instance keyword, you ensure that whenever any of the events listed in your start on condition is emitted, an instance of the job will be run. Therefore, if all three events are emitted very close together in time, three jobs instances will now be run.

See the Instance section for further details.

11.15   Run a Job Before Another Job

If you wish to run a particular job before some other job, simply make your jobs start on condition specify the starting(7) event. Since the starting(7) event is emitted just before the job in question starts, this provides the behaviour you want since your job will be run first.

For example, assuming your job is called job-B and you want it to start before job-A, in /etc/init/job-B.conf you would specify:

start on starting job-A

11.16   Run a Job After Another Job

If you have a job you wish to run after job "job-A", your start on condition would need to make use of the stopped(7) event like this:

start on stopped job-A

11.17   Run a Job Once After Some Other Job Ends

Imagine a job configuration file myjob.conf such as the following which might result in a job which is restarted a number of times:

start on event-A

  # do something
end script

Is it possible to run a job only once after job myjob ends? Yes if you create a job configuration file myjob-sync.conf such as:

start on stopped myjob and event-B

  # do something
end script

Now, when event-A is emitted, job myjob will start and if and when job myjob finishes and event event-B is emitted, job myjob-sync will be run.

However, crucially, even if job myjob is restarted, the myjob-sync job will not be restarted.

11.18   Run a Job Before Another Job and Stop it After that Job Stops

If you have a job you wish to be running before job "job-A" starts, but which you want to stop as soon as job-A stops:

start on starting job-A
stop on stopped job-A

11.19   Run a Job Only If Another Job Succeeds

To have a job start only when job-A succeeds, use the $RESULT variable from the stopped(7) event like this:

start on stopped job-A RESULT=ok

11.20   Run a Job Only If Another Job Fails

To have a job start only when job-A fails, use the $RESULT variable from the stopped(7) event like this:

start on stopped job-A RESULT=failed

Note that you could also specify this condition as:

start on stopped job-A RESULT!=ok

11.21   Run a Job Only If One Job Succeeds and Another Fails

This would be a strange scenario to want, but it is quite easy to specify. Assuming we want a job to start only if job-A succeeds and if job-B fails:

start on stopped job-A RESULT=ok and stopped job-B RESULT=failed

11.22   Run a Job If Another Job Exits with a particular Exit Code

Imagine you have a database server process that exits with a particular exit code (say 7) to denote that it needs some sort of cleanup process to be run before it can be re-started. To handle this you could create /etc/init/mydb-cleanup.conf with a start on condition like this:

start on stopped mydb EXIT_STATUS=7

  # handle cleanup...

  # assuming the cleanup was successful, restart the server
  start mydb
end script

11.23   Detect if Any Job Fails

To "monitor" all jobs for failures, you could either create a job that checks specifically for a single job failure (see Run a Job If Another Job Exits with a particular Exit Code), but you could just as easily detect if any job has failed as follows:

start on stopped RESULT=failed

Since this start on condition does not specify the Job to match against, it will match all jobs. You can then perform condition processing:

    if [ -n "$EXIT_STATUS" ];
        str="with exit status $EXIT_STATUS"
        str="due to signal $EXIT_SIGNAL"

    logger "Upstart Job $JOB (instance '$INSTANCE', process $PROCESS) failed $str"

    case "$JOB" in



end script

Note that $PROCESS above is not the PID, it is the name of the job process type (such as main or pre-start). See stopped(7) for further details.

11.24   Use Details of a Failed Job from Another Job

Although you cannot see the exact environment another job ran in, you can access some details. For example, if your job specified /etc/init/job-B.conf as:

start on stopped job-A RESULT=fail

  exec 1>>/tmp/log.file
  echo "Environment of job $JOB was:"
end script

The file /tmp/log.file might contain something like this:


Here, job-B can see that:

  • job-A exited in its "main" process. This is a special name for the script section. All other script sections are named as expected. For example, if the pre-start section had failed, the PROCESS variable would be set to pre-start, and if in post-stop, the variable would have been set to post-stop.
  • job-A exited with exit code 7.
  • job-A only had 1 instance (since the INSTANCE variable is set to the null value.
  • job-A ran in the root ("/") directory.
  • UPSTART_JOB is the name of the job running the script (ie job-B).
  • JOB is the name of the job that we are starting on (here job-A).
  • UPSTART_EVENTS is a list of the events that caused UPSTART_JOB (ie job-B) to start. Here, the event is starting(7) showing that job-B started as a result of job-A being sent the stopped(7) event.

11.25   Stop a Job when Another Job Starts

If we wish job-A to stop when job-B starts, specify the following in /etc/init/job-A.conf:

stop on starting job-B

11.25.1   Simple Mutual Exclusion

It is possible to create two jobs which will be "toggled" such that when job-A is running, job-B will be stopped and vice versa. This provides a simple mutually exclusive environment. Here is the job configuration file for job-A:

# /etc/init/job-A.conf
start on stopped job-B

  # do something when job-B is stopped
end script

And job-B:

# /etc/init/job-B.conf
start on stopped job-A

  # do something when job-A is stopped
end script

Finally, start one of the jobs:

# start job-A


  • when job-A is running, job-B will be stopped.
  • when job-B is running, job-A will be stopped.

Note though that attempting to have more than two jobs using such a scheme will not work. However, you can use the technique described in the Synchronisation section to achieve the same goal.

11.26   Run a Job Periodically

This cannot currently be handled by Upstart directly. However, the "Temporal Events" feature is being worked on now will address this.

Until Temporal Events are available you should either use cron(8), or something like:

# /etc/init/timer.conf

instance $JOB_TO_RUN

  for var in SLEEP JOB_TO_RUN
    eval val=\${$var}
    if [ -z "$val" ]
      logger -t $0 "ERROR: variable $var not specified"
      exit 1

  eval _sleep=\${SLEEP}
  eval _job=\${JOB_TO_RUN}

  while [ 1 ]
    stop  $_job || true
    sleep $_sleep
    start $_job || true
end script

Note well the contents of the while loop. We ensure that the commands that might fail are converted into expressions guaranteed to pass. If we did not do this, timer.conf would fail, which would be undesirable. Note too the use of instance to allow more than one instance of the timer job to be running at any one time.

11.27   Restart a job on a Particular Event

To restart a job when a particular event is emitted requires two jobs. First the main job:

start on something

exec /sbin/some-command

Then a helper job to perform the restart:

start on my-special-event

exec restart main-job

Now, when the my-special-event event is emitted, the main job will be restarted.

11.28   Migration from System V initialization scripts

With SysV init scripts, the Administrator decides the order that jobs are started in by assigning numeric values to each service. Such a system is simple, but non-optimal since:

  • The SysV init system runs each job sequentially.

    This disallows running jobs in parallel, to make full use of system resources. Due to the limited nature of the SysV system, many SysV services put services that take a long time to start into the background to give the illusion that the boot is progressing quickly. However, this makes it difficult for Administrators to know if a required service is running by the time their later service starts.

  • The Administrator cannot know the best order to run jobs in.

    Since the only meta information encoded for services is a numeric value used purely for ordering jobs, the system cannot optimize the services since it knows nothing about the requirements for each job.

In summary, the SysV init system is designed to be easy for the Administrator to use, not easy for the system to optimize.

In order to migrate a service from SysV to Upstart, it is necessary to change your mindset somewhat. Rather than trying to decide which two services to "slot" your service between, you need to consider the conditions that your service needs before it can legitimately be started.

So, if you wished to add a new service that traditionally started before cron(8) or atd(8) you do not need to change the configuration files cron.conf or atd.conf. You can "insert" your new service by specifying a simple:

# /etc/init/my-service.conf
start on (starting cron or starting atd)

In English, this says,

"start the "my-service" service just before either the cron or the atd services start".

Whether crond or atd actually start first is not a concern for my-service: Upstart ensures that the my-service service will be started before either of them. Even if cron normally starts before atd but for some reason one day atd starts first, Upstart will ensure that my-service will be started before atd.

Note therefore that introducing a new service should not generally require existing job configuration files to be updated.

11.29   How to Establish a Jobs start on and stop on Conditions

How do you establish what values you should specify for a jobs start on and stop on conditions?

11.29.1   Determining the start on Condition ( , )

So you have created a Job Configuration File for your Service Job. You have checked the expect stanza is correct and you've even enabled respawn.

But how do you determine the correct "start on" condition? Actually, this is almost a trick question since there are potentially many "correct" answers; it depends on the application and how sensitive it is to the environment it runs in. There are many potential start on conditions - it is your job to determine the most efficient and effective one. This section attempts to give some advice and guidelines on choosing a suitable condition, and explaining how to test your choice for correctness. However, note that each job requires a specific and possibly unique set of conditions to run.   Standard Idioms

If your application isn't particularly needy, you may be able to use one of the standard idioms below:

  • To start your job as soon as possible:

    See Run a Job as Soon as Possible After Boot.

  • To start your job "as late as possible":

    See Run a Job When a User Logs in Graphically (ubuntu-specific).

  • If you want the job to start "around the time" (actually just after) the equivalent System-V job would run, specify:

    start on stopped rc
  • If you want your job to start after all filesystems are mounted, specify:

    start on filesystem
  • If you want your job to start when all network devices are active, specify:

    start on stopped networking

    Note that as of Ubuntu Oneiric, you could also say:

    start on static-network-up
  • If you want your job to start when a runlevel begins, specify:

    start on runlevel [2345]

    This is used by a lot of standard jobs and is a good starting place.   More Exotic start on Conditions

If your job more precise control over when your job starts, read carefully the upstart-events(7) manual page which summarizes all the "well-known" events you can rely upon on an Ubuntu system. These events provide a set of "hook points" which your job can make use of to simplify the job of specifying the start on condition.

The main question to ask yourself is, "what are the exact requirements for the job?". To help answer that question consider the following questions:

  • Does your application live in a standard local directory?

  • Does the application write any files to disk? (data files, log files, lock files, named sockets?) If so, which partition(s) does it need to write to?

  • Does the application read any files from disk? If so, which partitions do they live in? /etc? /var?

  • Do you want the application to start as early as possible, or as late as possible?

  • Does the application need to start before or after a service which might not be installed?

  • If the application needs access to a disk (it probably will), which partitions or mounts does it need? /etc? /var? /mnt/remote-system? Can it wait until all local partitions are mounted? Or does it need to wait for a particular remote filesystem to be mounted?

  • Should a particular set of services already be running when your job starts?

  • Should a particular set of services not be running when your job starts?

  • What runlevel (or runlevels) should your job run in?

  • Does your application require a network?

    • Does it need a local network (
    • Does it need IPv6?
    • Does it require a bridge network interface?
  • Should your service only start when a client network connection is initiated? If so, use the socket event (emitted by the upstart-socket-bridge). See the socket-event(7) man page for details.

  • Does your job require the services of some other system server?

  • Does your job access files over the network?

  • Does your application provide a D-Bus service which you want to start when some sequence of Upstart events are emitted?

    If so, use the D-Bus service activation facility.

This list can be summarized as:

What are the precise conditions your job needs before it can be started successfully?

And yes, you really do need to be able to answer all the questions above before you can know that you have chosen the correct start on condition. This might sound daunting, but consider:

  • Upstart needs to know this information to allow your application to run at the correct point.
  • By devoting some time to understanding your applications requirements, you will allow the system to run as efficiently as possible.   udev conditions

To identify a start on condition making use of udev events, first you need to know which udev subsystem is appropriate. See upstart-udev-bridge for details.

Having identified the subsystem, follow the steps below:

  1. Create a job that displays all udev variables set for a particular udev subsystem.

    In the example below, we're consider at the tty subsystem, so modify to taste:

    start on tty-device-added
    exec env
  2. Boot your system and look at the relevant log file for the job.

    For example look at /var/log/upstart/myjob.log to see which udev variables are set for your chosen udev subsystem.

    If your version of Upstart does not have job logging, you'll need to redirect the output of env somewhere - refer to section See the Environment a Job Runs In.

  3. Refine your start on condition accordingly.

    For example, you might change it to be something like:

    start on tty-device-added DEVNAME=*ttyS1

    to start the job when the /dev/ttS1 serial device becomes available.

11.29.2   Determining the stop on Condition ()

Recall from the Shutdown section that if no stop on condition is stopped, your job will be killed at some (random) point at system shutdown. If you need your job to stop at a particular point in the shutdown sequence, you must specify a suitable stop on condition.

Shut down is not as event rich as startup. A common idiom is to specify your stop on as:

stop on runlevel [016]

This ensures the job will be stopped on shutdown, when switching to single-user mode and on reboot.

The next most common is to stop your job either before or after some other job stops:

  • To stop a job just before a particular job has started to stop:

    # stop your job "just before" job 'some-job' ends
    stop on stopping some-job

    See also Run a Job Before Another Job.

  • To stop a job immediately after a particular job has stopped:

    # stop your job "just after" job 'some-job' has ended
    stop on stopped some-job

    See also Run a Job After Another Job.

Other questions relating to other stanzas:

  • What should happen if your job fails to start?
  • What should happen if your job fails after some period of time?
  • Do you want Upstart to restart the job if it exits? If so, use the respawn stanza.
  • Does your job use non-standard exit codes to denote success and failure? If so, use the normal exit stanza.
  • Is your job a daemon? If so, how many times does it call fork(2)?

11.29.3   Final Words of Advice

If your start on or stop on conditions are becoming complex (referencing more than 2 or maybe 3 events), you should consider your strategy carefully since there is probably an easier way to achieve your goal by specifying some more appropriate event. See the upstart-events(7) manual page for ideas.

Also, review the conditions from standard job configuration files on your system. However, it is inadvisable to make use of conditions you do not fully understand.

11.30   Guarantee that a job will only run once

If you have a job which must only be run once, but which depends on multiple conditions, the naive approach won't necessarily work:

start on (A or B)

If event 'A' is emitted, the task will run. But assuming the task has completed and event 'B' is then emitted, the task will run again.

11.30.1   Method 1

A better approach is as follows:

  1. Create separate job configuration files for each condition you want your job to start on:

    # /etc/init/got-A.conf
    # job that will "run forever" when event A is emitted
    start on A
    # /etc/init/got-B.conf
    # job that will "run forever" when event B is emitted
    start on B
  2. Create a job which starts on either of the got-A or got-B jobs starting:

    # /etc/init/only-run-once.conf
    start on (starting got-A or starting got-B)

Now, job "only-run-once" will start only once since jobs "got-A" and "got-B" can only be started once themselves since:

  • they do not specify the instance stanza to allow multiple instances of the jobs.
  • if either job starts, that job will run forever.
  • none of the jobs have a stop on stanza.

11.30.2   Method 2

Change your start on condition to include the startup event:

start on startup and (A or B)

11.31   Stop a Job That is About to Start

Upstart will start a job when its "start on" condition becomes true.

Although somewhat unusual, it is quite possible to stop a job from starting when Upstart tries to start it:

start on starting job-A

  stop $JOB
end script

11.32   Stop a Job That is About to Start From Within That Job

You can in fact stop a job that Upstart has decided it needs to start from within that job:

pre-start script
end script

This is actually just an alias for:

pre-start script
end script

Of course, you could set the pre-start using the Override Files facility.

11.33   Stop a Job from Running if its Configuration file has not been Created/Modified

Use a pre-start stanza to check for required application conditions. If these are not met, call:

exit 0

This will cause the job to stop successfully before the main script or exec stanza (which would run your application/daemon) is started.

In particular, see the Ubuntu-specific example

11.34   Stop a Job When Some Other Job is about to Start

Here, we create /etc/init/job-C.conf which will stop job-B when job-A is about to start:

start on starting job-A

  stop job-B
end script

11.35   Start a Job when a Particular Filesystem is About to be Mounted

Here, we start a job when the /apps mountpoint is mounted read-only as an NFS-v4 filesystem:

start on mounting TYPE=nfs4 MOUNTPOINT=/apps OPTION=ro

Here's another example:

start on mounted MOUNTPOINT=/var/run TYPE=tmpfs

Another example where a job would be started when any non-virtual filesystem is mounted:

start on mounted DEVICE=[/UL]*

The use of the $DEVICE variable is interesting. It is used here to specify succinctly any device that:

  • is a real device (starts with "/" (to denote a normal "/dev/..." mount)).
  • is a device specified by its filesystem:
    • label (starts with "L" (to denote a "LABEL=" mount)).
    • UUID (starts with "U" (to denote a "UUID=" mount)).

Another example where a job is started when a non-root filesystem is mounted:

start on mounting MOUNTPOINT!=/ TYPE!=swap

11.36   Start a Job when a Device is Hot-Plugged

Hot-plug kernel events create udev(7) events under Linux and Upstart events are created from udev events by the upstart-udev-bridge(8).

Added to this the ifup and ifdown commands are run at boot when network devices are available for use.

11.36.1   To start a job when eth0 is added to the system

Note that the device is not yet be available for use):

start on net-device-added INTERFACE=eth0

See upstart-udev-bridge for more examples.

On an Ubuntu system, you can see which devices have been added by udev (which the upstart-udev-bridge is using) with this snippet:

$ awk 'BEGIN {RS=""; ORS="\n\n"}; /ACTION=add/ && /SUBSYSTEM=net/ { print; }' \
    /var/log/udev | grep ^INTERFACE= | cut -d= -f2 | sort -u

11.36.2   To start a job when eth0 is available

Here, the device is available for use:

start on net-device-up IFACE=eth0


  • It does not matter whether the eth0 interface has been configured statically, or if it is handled via DHCP, this event will always be emitted.

    See upstart-events(7) and file /var/log/udev for further details.

  • The "net-device-up" event sets the "IFACE" variable whereas the net-device-added event sets the "INTERFACE" variable!

11.37   Stopping a Job if it Runs for Too Long

To stop a running job after a certain period of time, create a generic job configuration file like this:

# /etc/init/timeout.conf
stop on stopping JOB=$JOB_TO_WAIT_FOR
kill timeout 1

export TIMEOUT

  sleep $TIMEOUT
  initctl stop $JOB_TO_WAIT_FOR
end script

Now, you can control a job using a timeout:

start myjob
start timeout JOB_TO_WAIT_FOR=myjob TIMEOUT=5

This will start job myjob running and then wait for 5 seconds. If job "myjob" is still running after this period of time, the job will be stopped using the initctl(8) command. Note the stop on stanza which will cause the timeout job not to run if the job being waited for has already started to stop.

11.38   Run a Job When a File or Directory is Created/Deleted

As of Upstart 1.8, you can use the upstart-file-bridge.

If you are using an older version of Upstart, read on...

If you need to start a Job only when a certain file is created, you could create a generic job configuration file such as the following:

# /etc/init/wait_for_file.conf
instance $FILE_PATH
export   FILE_PATH

  while [ ! -e "$FILE_PATH" ]
    sleep 1

  initctl emit file FILE_PATH="$FILE_PATH"
end script

Having done this, you can now make use of it. To have another job start if say file /var/run/foo.dat gets created, you first need to create a job configuration file stating this:

# /etc/init/myapp.conf
start on file FILE_PATH=/var/run/foo.dat

  # ...
end script

Lastly, kick of the process by starting an instance of wait_for_file:

start wait_for_file FILE_PATH=/var/run/foo.dat

Now, when file /var/run/foo.dat is created, the following will happen:

  1. The myapp job will emit the file event, passing the path of the file which you just specified in that events environment.
  2. Upstart will see that the start on condition for the myapp job configuration file is satisfied.
  3. Upstart will create a myapp job, and start it.

You can modify this strategy slightly to run a job when a file is:

  • modified
  • deleted
  • contains certain content
  • et cetera

See test(1), or your shells documentation for available file tests.

Note that this is very simplistic. A better approach would be to use inotify(7).

11.39   Run a Job Each Time a Condition is True

This is the default way Upstart works when you have defined a task:

# /etc/init/myjob.conf
exec /some/program
start on (A or B)

Job "myjob" will run every time either event 'A' or event 'B' are emitted. However, there is a corner condition: if event 'A' has been emitted and the task is currently running when event 'B' is emitted, job "myjob" will not be run. To avoid this situation, use instances:

# /etc/init/myjob2.conf
exec /some/program
start on (A or B)

Now, as long variable $SOME_VARIABLE is defined with a unique value each time either event 'A' or 'B' is emitted, Upstart will run job "myjob2" multiple times.

11.40   Run a Job When a Particular Runlevel is Entered and Left

To run a job when a particular runlevel is entered and also run it when that same runlevel is left, you could specify:

start on runlevel RUNLEVEL=5 or runlevel PREVLEVEL=5

See runlevel(7) and the Runlevels section for more details.

11.41   Pass State Between Job Processes

Assume you have a job configuration file like this:

pre-start script
   # ...
end script

exec /bin/some-program $ARG

How can you get the pre-start script section to set $ARG and have the "main" section use that value in the "exec" stanza? This isn't as easy as you might imagine for the simple reason that Upstart runs each script and exec section in a new process. As such, by the time Upstart gets to the exec stanza the process spawned to handle the pre-start script section has now ended. This implies they cannot communicate directly. However, there are ways to send information from one section to another...

One method to achieve the required goal is as follows:

# set a variable which is the name of a file this job will use
# to pass information between script sections.
env ARG_FILE="/var/myapp/myapp.dat"

# make the variable accessible to all script sections (ie sub-shells)
export ARG_FILE

pre-start script
   # decide upon arguments and write them to
   # $ARG_FILE, which is available in this sub-shell.
end script

  # read back the contents of the arguments file
  # and pass the values to the program to run.
  ARGS="$(cat $ARG_FILE)"

  # clean up
  rm -f $ARG_FILE || true

  exec /bin/some-program $ARGS
end script

However, as of Upstart 1.7, this is now possible (for Session Jobs only!) by using the initctl set-env command. For example:

pre-start script
  # modify the running jobs job environment table
  # such that when the 'exec' stanza is executed, Upstart will apply
  # all variables in this table to that job process.
  initctl set-env ARG=foo
end script

exec /bin/some-program $ARG

11.42   Pass State From Job Configuration File to a Script Section

To pass a value from a job configuration file to one of its script sections, simply use the env stanza:

env CONF_FILE=/etc/myapp/myapp.cfg

  exec /bin/myapp -c $CONF_FILE
end script

This example is a little pointless, but the following slightly modified example is much more useful:

start on an-event
export CONF_FILE

  exec /bin/myapp -c $CONF_FILE
end script

By dropping the use of the env stanza we can now pass the value in via an event:

# initctl emit an-event CONF_FILE=/etc/myapp/myapp.cfg

This is potentially much more useful since the value passed into myapp.conf can be varied without having to modify the job configuration file.

11.43   Run a Job as a Different User

11.43.2   Changing User

Some daemons start running as the super-user and then internally arrange to drop their privilege level to some other (less privileged) user. However, some daemons do not need to do this: they never need root privileges so can be invoked as a non-root user.

How do you run a "system job" but have it run as a non-root user then? As of Upstart 1.4, Upstart has the ability to run a System Job as a specified user using the setuid and setgid stanzas.

However, if you are not using Upstart 1.4, it is easy to accomplish the required goal. There are a couple of methods you can use. The recommended method for Debian and Ubuntu systems is to use the helper utility start-stop-daemon(8) like this:

exec start-stop-daemon --start -c myuser --exec command

The advantage of using start-stop-daemon(8) is that it simply changes the user and group the command is run as. This also has an advantage over su(1) in that su(1) must fork to be able to hold its PAM session open, and so is harder for upstart to track, whereas start-stop-daemon(8) will simply exec the given command after changing the uid/gid.

Another potential issue to be aware of is that start-stop-daemon does not impose PAM ("Pluggable Authentication Module") limits to the process it starts. Such limits can be set using the appropriate Upstart stanzas, you just cannot specify the limits via PAMs limits.conf(5).

Of course, you may want PAM restrictions in place, in which case you should either use su(1) or sudo(8), both of which are linked to the PAM libraries.

The general advice is NOT to use su(1) or sudo(8) though since PAM restrictions really not appropriate for system services. For example, PAM will make a wtmp(5) entry every time su(1) or sudo(8) are called and those records are not appropriate for system services.

If you want to use su(1) or sudo(8), the examples below show you how.

Using su(1):

exec su -s /bin/sh -c command $user

Note that although you could simplify the above to the following, it is not recommended since if user "$user" is a system account with a shell specified as /bin/false, the job will not run the specified command: it will fail due to /bin/false returning "1":

exec su -c command $user

The job will silently fail if user "$user" is a system account with a shell specified as /bin/false.

To avoid the fork(2) caused by the shell being spawned, you could instead specify:

exec su -s /bin/sh -c 'exec "$0" "$@"' $user -- /path/to/command --arg1=foo -b wibble

This technique is particularly useful if your job is a Service Job that makes use of expect.

A basic example using sudo(8):

exec sudo -u $user command

11.44   Disabling a Job from Automatically Starting

With Upstart 0.6.7, to stop Upstart automatically starting a job, you can either:

  • Rename the job configuration file such that it does not end with ".conf".
  • Edit the job configuration file and comment out the "start on" stanza using a leading '#'.

To re-enable the job, just undo the change.

11.44.1   Override Files

With Upstart 1.3, you can make use of an "override file" and the manual stanza to achieve the same result in a simpler manner [31]:

# echo "manual" >> /etc/init/myjob.override

Note that you could achieve the same effect by doing this:

# echo "manual" >> /etc/init/myjob.conf

However, using the override facility means you can leave the original job configuration file untouched.

To revert to the original behaviour, either delete or rename the override file (or remove the manual stanza from your ".conf" file).

For Session Jobs, note that if an override already exists "higher" up the search path, only that override file will apply: you cannot override an override file.   Change a Jobs Start/Stop Conditions

Override files allow you to modify the way in which a job starts and stop by modifying the start on and stop on conditions.

Maybe you are writing an application packaged for Upstart and you need to work around some known bugs, or maybe the system you are installing the package on is somehow non-standard and needs a special configuration. Again, override files can help. If your package has a job configuration file with a start on condition such as:

start on event-A

You can change that trivially with an override:

echo "start on (event-A and event-B)" >> /etc/init/myjob.override

Now the job will only start when both the event-A and event-B events are emitted. Again, to revert the behaviour back to starting when only event-A is emitted, just delete the override file.   Adding Stanzas that are Not Present in the .conf File

You can specify stanzas in the override file that are not in the original job configuration file and have Upstart use these values too. For example, imagine you have a daemon process that opens and writes to stderr with debug information if it sees a special "magic file":

$ cat >> /etc/init/my-daemon.override << EOT
console output
pre-start exec touch /var/run/my-daemon.magic

Note: We assume here that the corresponding /etc/init/my-daemon.conf file does not already specify a pre-start since if it did, our override file would replace it.   Separating Variables from the Job

Imagine you have a service which requires a number of variables to be defined. However, those variables vary depending on the system you intend to deploy the job to.

A way to resolve this problem is to use a static job coupled with an override file which you can generate for each host.

For example, your static job (say foo.conf) could look like this:

start on ...

exec myapp --log-directory $LOGDIR --port $PORT --foo $FOO_VALUE

Then, you can create a foo.override for one particular system containing for example:

env LOGDIR=/var/log/myapp
env PORT=12345
env FOO_VALUE="bar"   Ensuring Customized Packages Upgrade Smoothly

If you make changes to a packages configuration files, the chances are that you have a Package Manager which will notice this fact when it needs to upgrade the package. It may then prompt you to establish whether you want to keep your customized configuration file, overwrite it with the new package versions configuration file, or some other possibilities.

If you use override files to encode any modifications you make to a packages job configuration, you can avoid this issue (at least for job configuration) and ensure your configuration is always used (since the job configuration files itself never needs to be changed).   Caveat Emptor

If override files are used, you should employ other facilities to detect when an underlying job configuration file has actually changed.

This would be necessary for example should a new job configuration file be installed that fixed an important bug: if your override file overrides the particular configuration option that the new version resolved a issue with, your system may still exhibit the bug since you have overridden the (fixed) configuration option and thus may have unwittingly "undone" the fix. This problem is certainly not unique to Upstart override files, but it is worth considering.

11.45   Jobs that "Run Forever"

To create a job that runs continuously from the time it is manually started(7) until the time it is manually stopped(7), create a job configuration file without any process definition (exec and script) or event definition (start on for example) stanzas:

# /etc/init/runforever.conf
description "job that runs until stopped manually"

This job can only be started by the administrator running:

# start runforever

The status of this job will now be "start/running" until the administrator subsequently runs:

# stop runforever

These "Abstract Job" types have other uses as covered in other parts of this document. See for example Synchronisation.

11.46   Run a Java Application

Running a Java application is no different to any other, but Java suffers from the inability to switch users without extra helper classes.

If your Java daemon needs to run as a different user and you are running Upstart 1.4, you can use the setuid and setgid stanzas.

However, if you are using an older version, you will have to use a facility such as su(1). Also, you may wish to define some variables to simplify the invocation:

env ROOT_DIR=/apps/myapp
env HTTP_PORT=8080
env USER=java_user
env JAVA_HOME=/usr/lib/jvm/java-6-openjdk
env JVM_OPTIONS="-Xms64m -Xmx256m"
env APP_OPTIONS="--httpPort=$HTTP_PORT"
env LOGFILE=/var/log/myapp.log

  exec su -c "$JAVA_HOME/bin/java $JVM_OPTIONS \
    -jar $ROOT_DIR/myjar.jar $APP_OPTIONS > $LOGFILE 2>&1" $USER
end script

You should read the Changing User section section before using this technique though.

11.46.1   Alternative Method

Here is how you might run a Java application which calls fork(2) some number of times:

exec start-stop-daemon --start --exec $JAVA_HOME/bin/java \
  -- $JAVA_OPTS -jar $SOMEWHERE/file.war

Again, you should read the Changing User section section before using this technique.

11.47   Ensure a Directory Exists Before Starting a Job

This is a good use of the pre-start stanza:

env DIR=/var/run/myapp
env USER=myuser
env GROUP=mygroup
env PERMS=0755

pre-start script
  mkdir $DIR              || true
  chmod $PERMS $DIR       || true
  chown $USER:$GROUP $DIR || true
end script

11.48   Run a GUI Application

To have Upstart start a GUI application, you first need to ensure that the user who will be running it has access to the X display. This is achieved using the xhost command.

Once the user has access, the method is the same as usual:

env DISPLAY=:0.0
exec xclock -update 1

11.49   Run an Application through GNU Screen

If you want Upstart to create a GNU Screen (or Byobu) session to run your application in, this is equally simple:

exec su myuser -c "screen -D -m -S MYAPP java -jar MyApp.jar"

11.50   Run Upstart in a chroot Environment

11.50.1   chroot Workaround for Older Versions of Upstart ( , )

Older versions of Upstart jobs cannot be started in a chroot(2) environment [21] because Upstart acts as a service supervisor, and processes within the chroot are unable to communicate with the Upstart running outside of the chroot. This will cause some packages that have been converted to use Upstart jobs instead of init scripts to fail to upgrade within a chroot.

Users are advised to configure their chroots with /sbin/initctl pointing to /bin/true, with the following commands run within the chroot:

dpkg-divert --local --rename --add /sbin/initctl
ln -s /bin/true /sbin/initctl

11.50.2   chroots in Ubuntu Natty ()

The version of Upstart in Ubuntu Natty now has full chroot(2) support. This means that if initctl is run as user root from within a chroot the Upstart init daemon (outside the chroot) will honour requests from within the chroot to manipulate jobs within the chroot.

What all this means is that you no longer need to use dpkg-divert and can control chroot jobs from within the chroot environment exactly as you would control jobs outside a chroot environment. There are a number of caveats and notes to consider though:

  • Within the chroot, only jobs within the chroot are visible

  • Within the chroot, only jobs within the chroot can be manipulated.

  • It is only possible to view and control such chroot jobs from within the chroot.

    That is to say, the "outer" system cannot manipulate jobs within the chroot.

  • Due to the design of this feature, Upstart will not be able to detect changes to job configuration files within the chroot until a process within the chroot has either manipulated a job, or listed one or more jobs.

  • Chroot support can be disabled at boot by passing the "--no-sessions" option on the Grub kernel command-line.

    See Add --verbose or --debug to the kernel command-line for details of how to add values to the grub kernel command-line.

    If chroots are disabled, running Upstart commands within a chroot will affect jobs outside the chroot only.

    Note that "--no-sessions" was removed in Upstart 1.13 since the default in newer versions of Upstart is to run with chroot sessions diabled. To enable chroot support, it is now necessary to specify "--chroot-sessions".

  • If a job is run in a chroot environment (such as provided by schroot(1)), exiting the chroot will kill the job.

11.51   Record all Jobs and Events which Emit an Event

For example, if you want to record all jobs which emit a started event:

# /etc/init/debug.conf
start on started
  exec 1>>/tmp/log.file
  echo "$0:$$:`date`:got called. Environment of job $JOB was:"
end script

You could also log details of all jobs (except the debug job itself) which are affected by the main events:

# /etc/init/debug.conf
start on ( starting JOB!=debug \
  or started JOB!=debug \
  or stopping JOB!=debug \
  or stopped JOB!=debug )
  exec 1>>/tmp/log.file
  echo -n "$UPSTART_JOB/$UPSTART_INSTANCE ($0):$$:`date`:"
  echo    "Job $JOB/$INSTANCE $UPSTART_EVENTS. Environment was:"
end script

Note that the $UPSTART_JOB and $UPSTART_INSTANCE environment variables refer to the debug job itself, whereas $JOB and $INSTANCE refer to the job which the debug job is triggered by.

11.52   Integrating your New Application with Upstart

Integrating your application into Upstart is actually very simple. However, you need to remember that Upstart is NOT "System V" (aka "SysV"), so you need to think in a different way.

With SysV you slot your service script between other service scripts by specifying a startup number. The SysV init system then runs each script in numerical order. This is very simple to understand and use, but highly inefficient in practical terms since it means the boot cannot be parallelised and thus cannot be optimized.

11.53   Block Another Job Until Yours has Started

It is common that a particular piece of software, when installed, will need to be started before another. The logical conclusion is to use the 'starting' event of the other job:

start on starting foo

This will indeed, block foo from starting until our job has started.

But what if we have multiple events that we need to delay:

start on starting foo or starting network-services

This would seem to make sense. However, if we have a time-line like this:

starting foo
starting our job
starting network-services
started network-services

Network-services will actually NOT be blocked. This is because upstart only blocks an event if that event causes change in the goal of the service. So, we need to make sure upstart waits every time. This can be done by using a "wait job":

# myjob-wait
start on starting foo or starting network-services
stop on started myjob or stopped myjob
instance $JOB
normal exit 2
  status myjob | grep -q 'start/running' && exit 0
  start myjob || :
  sleep 3600
end script

This is a bit of a hack to get around the lack of state awareness in Upstart. Eventually this should be built in to upstart. The job above will create an instance for each JOB that causes it to start. It will try and check to see if it's already running, and if so, let the blocked job go with exit 0. If it's not running, it will set the ball in motion for it to start. By doing this, we make it very likely that the stopped or started event for myjob will be emitted (the only thing that will prevent this, is a script line in 'myjob' that runs 'stop'). Because we know we will get one of those start or stopped events, we can just sleep for an hour waiting for upstart to kill us when the event happens.

11.54   Controlling Upstart using D-Bus

Upstart contains its own D-Bus server which means that initctl and any other D-Bus application can control Upstart. The examples below use dbus-send, but any of the D-Bus bindings could be used.

11.54.1   Query Version of Upstart

To emulate initctl version, run:

$ dbus-send --system --print-reply --dest=com.ubuntu.Upstart /com/ubuntu/Upstart org.freedesktop.DBus.Properties.Get string:com.ubuntu.Upstart0_6 string:version

Note: this is querying the version of /sbin/init, not the version of initctl. For the latter, see initctl version.

11.54.2   Query Log Priority

To emulate initctl log-priority and show the current log priority, run:

$ dbus-send --system --print-reply --dest=com.ubuntu.Upstart /com/ubuntu/Upstart org.freedesktop.DBus.Properties.Get string:com.ubuntu.Upstart0_6 string:log_priority

11.54.3   Set Log Priority

To emulate initctl log-priority and set a new log priority, run:

$ priority=debug
$ sudo dbus-send --system --print-reply --dest=com.ubuntu.Upstart /com/ubuntu/Upstart org.freedesktop.DBus.Properties.Set string:com.ubuntu.Upstart0_6 string:log_priority variant:string:$priority

11.54.4   List all Jobs via D-Bus

To emulate initctl list, run:

$ dbus-send --system --print-reply --dest=com.ubuntu.Upstart /com/ubuntu/Upstart com.ubuntu.Upstart0_6.GetAllJobs

11.54.5   Get Status of Job via D-Bus

To emulate initctl status, run:

$ job=myjob
$ dbus-send --system --print-reply --dest=com.ubuntu.Upstart /com/ubuntu/Upstart/jobs/${job}/_ org.freedesktop.DBus.Properties.GetAll string:''

Note that this will return information on all running job instances of myjob.

11.54.6   Emit an Event

To emulate initctl emit, run:

$ event=foo
$ sudo dbus-send --system --print-reply --dest=com.ubuntu.Upstart /com/ubuntu/Upstart com.ubuntu.Upstart0_6.EmitEvent string:$event array:string: boolean:true

To emulate initctl emit --no-wait <event> A=B c='hello world' D=123.456, run:

$ event=foo
$ sudo dbus-send --system --print-reply --dest=com.ubuntu.Upstart /com/ubuntu/Upstart com.ubuntu.Upstart0_6.EmitEvent string:$event array:string:'A=B',"C='hello world'",D=123.456 boolean:false

11.54.7   Get Jobs start on and stop on Conditions via D-Bus

To show a jobs start on and stop on conditions:

$ job=cron
$ for condition in start_on stop_on
> do
>     dbus-send --system --print-reply --dest=com.ubuntu.Upstart /com/ubuntu/Upstart/jobs/$job org.freedesktop.DBus.Properties.Get string:com.ubuntu.Upstart0_6.Job string:$condition
> done

If you have a job with a start on condition like this:

start on (starting foo A=B or (stopping bar C=D and (stopped baz E=F G=H I=J or foo)))

... a dbus-send(1) query like the one above for start on will return an "array of arrays of strings":

method return sender=:1.629 -> dest=:1.630 reply_serial=2
   variant       array [
         array [
            string "starting"
            string "foo"
            string "A=B"
         array [
            string "stopping"
            string "bar"
            string "C=D"
         array [
            string "stopped"
            string "baz"
            string "E=F"
            string "G=H"
            string "I=J"
         array [
            string "foo"
         array [
            string "/OR"
         array [
            string "/AND"
         array [
            string "/OR"

This will require a little massaging. Every inner array entry represents one of the following:

  • an Event
  • an operator ("and" or "or")

For event arrays, the first element is the event name and subsequent elements represent the events environment variables.

Note too that the entire start on expression has been encoded using Reverse Polish Notation (RPN) since this is a convenient format to represent the condition (particularly when you consider that they are represented internally as trees).

Normally, you don't need to get involved with RPN since initctl show-config converts the RPN back into the original form as specified in the Job Configuration file.

11.54.8   To Start a Job via D-Bus

To emulate initctl start, run:

# job=myjob
# dbus-send --system --print-reply --dest=com.ubuntu.Upstart /com/ubuntu/Upstart/jobs/${job} com.ubuntu.Upstart0_6.Job.Start array:string: boolean:true

Note that you must be root to manipulate system jobs.

11.54.9   To Stop a Job via D-Bus

To emulate initctl stop, run:

# job=myjob
# dbus-send --system --print-reply --dest=com.ubuntu.Upstart /com/ubuntu/Upstart/jobs/${job} com.ubuntu.Upstart0_6.Job.Stop array:string: boolean:true

Note that you must be root to manipulate system jobs.

11.54.10   To Restart a Job via D-Bus

To emulate initctl restart, run:

# job=myjob
# dbus-send --system --print-reply --dest=com.ubuntu.Upstart /com/ubuntu/Upstart/jobs/${job} com.ubuntu.Upstart0_6.Job.Restart array:string: boolean:true

Note that you must be root to manipulate system jobs.

11.55   Establish Blocking Job

Image you have just run the following command and it has "blocked" (appeared to hang):

# initctl emit event-A

The reason for the block is that the event-A event changes the goal of "some job", and until the goal has changed, the initctl command will block.

But which job is being slow to change goal? It is now possible to hone in on the problem using initctl show-config in a script such as this:


[ $# -ne 1 ] && { echo "ERROR: usage: $0 <event>"; exit 1; }

# obtain a list of jobs (removing instances)
initctl list | awk '{print $1}' | sort -u | while read job
  initctl show-config -e "$job" |\
    egrep "(start|stop) on \<event\>" >/dev/null 2>&1
  [ $? -eq 0 ] && echo $job

This will return a list of jobs, one per line. One of these will be the culprit. Having identified the problematic job, you can debug using techniques from the Debugging section.

11.56   Determine if a Job is Disabled

To determine if a job has been disabled from starting automatically:

$ job=foo
$ initctl show-config $job | grep -q "^  start on" && echo enabled || echo disabled

11.57   Visualising Jobs and Events

Use the initctl2dot(8) facility. See [29] for further details and examples.

11.58   Sourcing Files

You need to take care when "sourcing" a script or configuration file into a script section for a number of reasons. Suppose we have the following:

. /etc/default/myapp.cfg
. /etc/myapp/myapp.cfg
echo hello > /tmp/myapp.log
end script

Assume that file /etc/myapp/myapp.cfg does NOT exist.

11.58.1   Develop Scripts Using /bin/sh

Firstly, if you developed this script using the bash(1) shell, before you put it into a job configuration file), all would be well. However, as noted, Upstart runs all jobs with /bin/sh -e. What you will find is that if you run the script above under /bin/sh, in all likelihood the file will never be created since regardless of whether you specify "-e" or not, the dash(1) shell (which /bin/sh is linked to on Ubuntu systems) has different semantics when it comes to sourcing compared with /bin/bash.

Therefore, to avoid surprises later on:

  • Always develop your scripts using "/bin/sh -e".

  • Always code defensively.

    For example, it would be better to write the script above as:

      [ -f /etc/default/myapp.cfg ] && . /etc/default/myapp.cfg
      [ -f /etc/myapp/myapp.cfg ]   && . /etc/myapp/myapp.cfg
      echo hello > /tmp/myapp.log
    end script

Or maybe even like this to minimise mistakes:


  for file in $files
    [ -f "$file" ] && . "$file"
  echo hello > /tmp/myapp.log
end script

11.58.2   ureadahead

Most modern Linux systems attempt to optimise the boot experience by pre-loading files early on in the boot sequence. This allows hard disks can minimise expensive (slow) seek operations.

On Ubuntu, this job is accomplished using ureadahead(8), which was designed with both spinning hard disk and SSD drives in mind. However, if your job configuration files start reading files from all over the disk, you will be potentially slowing down the boot as the disk is then forced to seek across the filesystem, looking for your files.

The general advice is therefore to put your configuration variables inside the job configuration file itself where possible.

11.59   Determining How to Stop a Job with Multiple Running Instances

As explained in the initctl status section, a job that has multiple running instances will show the specific (unique) instance value within brackets:

$ initctl list | grep ^network-interface-security
network-interface-security (network-manager) start/running
network-interface-security (network-interface/eth0) start/running
network-interface-security (network-interface/lo) start/running
network-interface-security (networking) start/running

In the example output above there are four instances of the network-interface-security job running with the unique instances values of:

  • "network-manager"
  • "network-interface/eth0"
  • "network-interface/lo"
  • "networking"

So how do we stop one of these jobs? Lets try to work this out without looking at initctl(8) manual page:

# stop network-interface-security network-interface/eth0
stop: Env must be KEY=VALUE pairs

That clearly doesn't work. The problem is that we have provided the value to the instance variable, but we haven't named the instance variable that the given value corresponds to. But how do we establish the instance variable name?

There are 2 options:

  • look at the corresponding Job Configuration File.

    /etc/init/network-interface-security.conf in this example.

  • Use a trick to get Upstart to tell you the name:

    $ status network-interface-security
    status: Unknown parameter: JOB

    This shows us the name of the instance variable is "JOB".

We are now in a position to stop a particular instance of this job:

# stop network-interface-security JOB=network-interface/eth0
network-interface-security stop/waiting

The job instance has now been stopped. To prove it:

# status network-interface-security JOB=network-interface/eth0
status: Unknown instance: network-interface/eth0
# initctl list | grep ^network-interface-security | grep network-interface/eth0

11.60   Logging Boot and Shutdown Times

If you want to create a log of when your system starts and stops, you could do something like this:

start on filesystem or runlevel [06]

env log=/var/log/boot-times.log

  action=$(echo "$UPSTART_EVENTS" | grep -q filesystem && echo boot || echo shutdown)
  echo "`date`: $action" >> $log
end script

Note that you do not need to specify a stop on condition: you want this job to start both "around" the time of system startup (when the disks are writeable, hence the use of the filesystem event) and shutdown.

If you want a more accurate method, you would need to have a job start on startup. The slight issue here is that when Upstart emits that first event, there is no guarantee of writeable disks. However, this can be overcome using a bit of thought...

First, create a "record-boot-time.conf" job configuration file to record the time of the "boot" (initial Upstart event):

start on startup

exec initctl emit boot-time TIME=$(date '+%s')

This job emits an event containing a variable specifying the time in seconds since the Epoch.

Now, create a second "log-boot-time.conf" job configuration file to actually log the boot time:

start on boot-time and filesystem


  echo "system booted at $TIME" >>$log
end script

Since the "log-boot-time" job specifies the "booted" event emitted by the "record-boot-time" job, Upstart will retain knowledge of this event until it is able to run the second job. The "record-boot-time" job can then simply make use of the "TIME" variable set by the first job.

11.61   Running an Alternative Job on a tty

Here's a silly example of how to run a custom job on a particular tty. It asks the user to guess a random number. If after 3 attempts they fail to guess the correct number, the job ends. However, if they guess successfully, the are allowed to login. This won't win any scripting competitions, but you get the idea.

WARNING - DO NOT USE THIS ON A REAL SYSTEM unless you want to get hacked, or fired or both!:

# Get the user to guess the number. If they get it right, let them
# login.

start on runlevel [23]
stop on runlevel [!23]

env tty=tty9

# XXX: Ensure job is connected to the terminal device
console output

  # XXX: Ensure all standard streams are connected to the console
  exec 0</dev/$tty >/dev/$tty 2>&1
  trap '' INT TERM HUP
  RANDOM=$(dd if=/dev/urandom count=1 2>/dev/null|cksum|cut -f1 -d' ')
  answer=$(((RANDOM % 100) + 1))

  while [ $attempt -lt $max ]
    echo -n "Guess the number (1-100, attempt $attempt of $max): "
    read guess
    if [ "$guess" -eq "$answer" ]
      echo "Wrong"

  [ "$got" = 0 ] && stop

  exec /sbin/getty -8 38400 $tty
end script

The important lines are:

console output

... and:

exec 0</dev/$tty >/dev/$tty 2>&1

11.62   Delay Respawn of a Job

If a job specifies respawn, but you want to delay the respawn for some reason, simply use a post-stop stanza:

exec mydaemon
post-stop exec sleep 10

Now, every time mydaemon exits with a non-zero return code, the job will sleep for 10 seconds before Upstart restarts it.

For a real service, you would probably use a post-stop script stanza to perform a check rather than simply sleeping.

11.63   Allow a job to detect if it was stopped manually

A job can detect if it itself was stopped manually by an administrator job using the stop command, by virtue of the fact that the $UPSTART_EVENTS environment variable will not be set in the jobs environment (since no event caused the job to stop - an administrator intervened).

See Standard Environment Variables for details of environment variables Upstart sets.

11.64   Detect if a job stopped before reaching its respawn limit

In this scenario, assuming the job actually specifies the respawn stanza, the $PROCESS variable of the stopping` event (see Event) will be set to a value other than respawn - it will be set to the name of the particular job process type (pre-start, main, post-stop, et cetera) that failed.

Upstart will then automatically restart the job.

To create a job that reacts to a job stopping before it reaches its respawn limit:

start on stopped PROCESS!=respawn


  # Exit if $JOB has not specified the respawn stanza.
  # Unfortunately, Upstart does not provide a good way to do this
  # aside from using grep on the job file to look for the stanza.
  egrep "^ *\<respawn\> *$" /etc/init/${JOB}.conf && exit 0 || true

  echo "respawn job '$JOB' (instance '$INSTANCE') will be restarted"

end script


11.66   Detecting a job hitting its respawn limit

To create a watchdog job that can react to jobs which hit their respawn limit:

start on stopped RESULT="failed" PROCESS="respawn"

exec echo "ERROR: `date`: job '$JOB' (instance '$INSTANCE') hit respawn limit"

Note that unlike the job in Detect if a job stopped before reaching its respawn limit, we don't need to check for the respawn stanza since Upstart only sets PROCESS="respawn" value for a job that has specified the respawn.

Rather than just logging an error, the job could take any number of actions. For example, it could log a error, then manually restart the job, setting an extra variable so that the job itself could detect that it had failed and possibly fall back to some other configuration:


  exec echo "ERROR: `date`: job '$JOB' (instance '$INSTANCE') hit respawn limit - restarting"

  start "$JOB" RESTARTED_BY_WATCHDOG=1 || true

end script

11.67   Identifying jobs that may need a respawn stanza

By default, Upstart will not respawn a job - you need to specify the respawn stanza.

But, is there a way to determine whether you should add the respawn stanza to a job?

There are 2 main reasons for using the respawn stanza:

  • To handle jobs that legitimately exit and need to be restared.

    A good example of this category of jobs is getty. See /etc/init/tty1.conf on an Ubuntu system for an example.

  • To handle buggy applications / daemons

    Using respawn for this class of jobs may be a practical solution but must be seen as a temporary plaster / bandaid over a gaping wound - the cause of the crashes should be identified and the code fixed.

That said, to highlight jobs that could benefit from the addition of the respawn stanza, you could create a watcher job that specifies:

start on stopped RESULT=failed PROCESS!=respawn

  # ignore jobs that already specify respawn
  egrep "^ *\<respawn\> *$" /etc/init/${JOB}.conf && exit 0 || true

  echo "job '$JOB' may benefit from adding 'respawn'"

end script

Then, review the results and determine if it is appropriate to either fix the code or add respawn (possibly temorarily).

11.68   Creating a SystemV Service that Communicates with Upstart

There are occasions when you want to have a SystemV service start an Upstart job. However, you must take care as shown in the example below...

Image we create a SysV service as /etc/init.d/myservice. This service needs another service to be running but that other service is actually an Upstart job (/etc/init/myjob.conf).

The Upstart job specifies a start on condition of:

start on filesystem and static-network-up and myservice-server-running

So, job myjob will only start once all three of the events specified are emitted and the myservice-server-running event is being emitted by /etc/init.d/myservice like this:

initctl emit myservice-server-running

This all looks perfectly reasonable and in fact it is... generally.

However, consider what would happen if the package containing /etc/init.d/myservice happened to attempt to restart that service having installed it (to make sure it is running immediately after installation)...

  1. /etc/init.d/myservice is run.
  2. /etc/init.d/myservice calls "initctl emit myservice-server-running".
  3. Upstart emits the myservice-server-running event.

Nothing magical here yet. Or is there? Since job myjob will only be started when all three of the events specified in its start on condition are true, this job cannot yet be started. Why? Because the filesystem and static-network-up events have already been emitted early in the boot (see Ubuntu Well-Known Events (ubuntu-specific)).

What this means is that the job myjob will never start post boot if those two events it cares about have already been emitted. Any yet, the SysV job and the Upstart event combinations are perfectly valid on boot. Note too that because those two events will not be re-emitted, the initctl emit will block (appear to hang) since Upstart is waiting for those two events to be emitted.

The solution to this is very simple: make the SysV job only emit the event in question on boot:

# Only emit the event 'on boot' to ensure the SysV service
# does not "hang" (block) due to events the ``myjob`` job requires
# never being re-emitted post-boot. We do this by checking for one of
# Upstarts  standard environment variables which will only be run when
# the Upstart SysV compatibility system is running the SysV service in
# question.
[ -n "$UPSTART_JOB" ] && initctl emit myservice-server-running

A slightly different method is to emit a signal by running initctl with the --no-wait option like this:

[ -n "$UPSTART_JOB" ] && initctl emit --no-wait myservice-server-running

See Signals and Standard Environment Variables.

11.69   Running a job in a cgroup

As of Upstart 1.13, cgroups are supported. See cgroup.

For older versions of Upstart in versions of Ubuntu from Trusty onwards, you can make use of the provided Cgroup management daemon called cgmanager (available in a package of the same name). Upstart job processes can be cgroup-contained by making use of the cgm(1) utility available in the cgmanager-utils package. Note that each job process that needs to be cgroup-contained need to call ``cgm`` appropriately.


  • To run just the main (exec or script) job process in an existing cgroup:

    start on ...
      cgm movepid cpu an-existing-group $$
      exec myprog --arg1 "foo"
    end script
  • To run the pre-start, main and pre-stop job processes in a new memory cgroup with a modified memory limit:

    start on ...
    env cgroup_name="foo"
    env cgroup_controller="memory"
    pre-start script
      cgm create "$cgroup_controller" "$cgroup_name"
      cgm setvalue "$cgroup_controller" "$cgroup_name" limit_in_bytes 52428800
      cgm movepid "$cgroup_controller" "$cgroup_name" $$
      # ...
    end script
      cgm movepid "$cgroup_controller" "$cgroup_name" $$
      exec myprog --arg1 "foo"
    end script
    pre-stop script
      cgm movepid "$cgroup_controller" "$cgroup_name" $$
      # ...
    end script

See the cgm(1) manual page for further details.

12   Test Your Knowledge

12.1   Questions about start on

Consider the following start on condition:

start on startup or starting stopped or stopping started

Questions (answers provided in footnote links):

Question:Is this a legal condition?
Question:What standard Upstart tool could you use to help explain the expression?
Question:Explain the condition.
Question:How many times could this job be run assuming all other jobs on the system run exactly once?

Consider this start on condition:

start on not foo
Question:Is this a legal condition?
Question:What event will cause the job to start?
Question:What is foo in this context?
Question:How could you trigger this job to run using initctl emit?

12.2   General Questions

What is wrong with the following job configuration file?:

start on startup

  echo hello > /tmp/foo.log
end script

Answer: [9]

What is wrong with the following job configuration file?:

start on runlevel [2345]

env CONFIG=/etc/default/myapp

expect fork

  enabled=$(grep ENABLED=1 $CONFIG)
  [ -z "$enabled" ] && exit 0
end script

Answer: [10]

13   Common Problems

13.1   Cannot Start a Job

If you have just created or modified a job configuration file such as /etc/init/myjob.conf, but start gives the following error when you attempt to start it:

start: Unknown job: myjob

The likelihood is that the file contains a syntax error. The easiest way to establish if this is true is by running the init-checkconf command.

If you are wondering why the original error couldn't be more helpful, it is important to remember that the job control commands (start, stop and restart) and initctl communicate with Upstart over D-Bus. The problem here is that Upstart rejected the invalid myjob.conf, so attempting to control that job over D-Bus is nonsensical - the job does not exist.

13.2   Cannot stop a job

If start is hanging or seems to be behaving oddly, the chances are you have misspecified the expect stanza. See expect and How to Establish Fork Count.

13.3   Strange Error When Running start/stop/restart or initctl emit

If you attempt to run a job command, or emit an event and you get a D-Bus error like this:

$ start myjob
start: Rejected send message, 1 matched rules; type="method_call", sender=":1.58" (uid=1000 pid=5696 comm="start) interface="com.ubuntu.Upstart0_6.Job" member="Start" error name="(unset)" requested_reply=0 destination="com.ubuntu.Upstart" (uid=0 pid=1 comm="/sbin/init"))

The problem is caused by not running the command as root. To resolve it, either "su -" to root or use a facility such as sudo(8):

# start myjob
myjob start/running, process 1234

The reason for the very cryptic error is that the job control commands (start, stop and restart) and initctl communicate with Upstart over D-Bus.

13.4   The initctl command shows "the wrong PID"

The likelihood is that you have mis-specified the type of application you are running in the job configuration file. Since Upstart traces or follows fork(2) calls, it needs to know how many forks to expect. If your application forks once, specify the following in the job configuration file:

expect fork

However, if your application forks twice (which all daemon processes should do), specify:

expect daemon

See also Alternative Method.

13.6   Sometimes status shows PID, but other times does not

You may have noticed that when you start certain jobs manually using start, sometimes the output will show the PID of the process associated with that job. However, other times, no PID is shown. Why?

This behaviour is observed when the job runs to completion very quickly. If your system has minimal load the job will start and finish before the initctl status command has a chance to query its PID from Upstart. Whereas if your system is busy you may well see a PID displayed since Upstart was able to return the PID details to status before the job finished.

The behaviour is similar to the following shell code:

(sleep 0.01 &) ; ps -fU $USER | grep sleep | grep -v grep

It is unlikely that you will get any output from this command (since the sleep 0.01 command will run to completion before the grep(1) calls get a chance filter the ps(1) output. However, change the time for that subshell to run, and you will see the PID:

(sleep 5 &) ; ps -fU $USER | grep sleep | grep -v grep

See initctl status.

14   Upstart in Debian and Ubuntu ( , )

14.1   Packaging

Simply create you Job Configuration File in the package but rather than having the file end with .conf, ensure the suffix is .upstart such as:


debhelper(7) will then automatically install this job as /etc/init/${package}.conf.

As long as you do not override dh_install (override_dh_install or override_dh_auto_install in debian/rules), this will "just work" as debhelper(7) will automatically invoke dh_installinit(1) for you.

If you do override dh_install, ensure you invoke dh_installinit in debian/rules.

  • service(8) just "does the right thing".
  • update-rc.d(8) just "does the right thing".
  • invoke-rc.d(8) just "does the right thing".
  • Do not have your package call /etc/init.d/${package} directly since:
    • it is not necessary.
    • it assumes a particular init system.

14.3   Fun with Job Files ( , )

14.3.1   Identify Missing System Jobs ( , )

If somebody inadvertently deleted a job file, here's one way to establish which ones:

dpkg -l | grep ^ii | awk '{print $2}' | xargs dpkg -s | grep '^ /' |\
    awk '{print $1}' | grep "^/etc/init/" | while read file
        [ ! -f "$file" ] && echo "job '$file' missing"

It prints a list of missing job files, one per line.

14.3.2   Identify Modified System Jobs ( , )

To find out if any job files have been modified from their "pristine" installed state, run this:

dpkg-query -W -f='${Conffiles}\n' '*' | grep "^ /etc/init/" |\
    awk 'OFS=" "{print $2,$1}' | md5sum -c 2>/dev/null |\
    awk -F': ' '$2 !~ /OK/ {print $1}'

The script above prints the full path to all modified job files, one per line. In fact, it will print missing files as well as modified files.

14.3.3   Identify Non-Packaged System Jobs ( , )

To get a list of jobs that were created manually (in other words are not part of official packages), run:

# First, get a list of all Upstart jobs in all packages installed on
# your system, and write it to a data file.
dpkg-query -W -f='${Conffiles}\n' '*' | grep "^ /etc/init/" |\
    awk '{print $1}' | sed 's!/etc/init/!!g' > /tmp/upstart.dat

# Now list all jobs *NOT* in the data file from above
ls /etc/init/*.conf | sed 's!^/etc/init/!!g' |\
while read file
    grep -q "^${file}$" /tmp/upstart.dat || echo "/etc/init/$file"

14.3.4   Re-install all Packages with Missing or Modified System Job Files ( , )


Before trying this, make sure you understand the implications and that you have a full system backup!!

To forcibly re-install all packages which have either had their job files deleted or modified in some way, run the following to get back to a "pristine" state:

# Install required helper application
$ sudo apt-get install -y apt-file

# Get a list of modified or missing files
$ dpkg-query -W -f='${Conffiles}\n' '*' | grep /etc/init/ |\
    awk 'OFS=" "{print $2,$1}' | md5sum -c 2>/dev/null |\
    awk -F': ' '$2 !~ /OK/{print $1}' > /tmp/upstart.dat

$ cat /tmp/upstart.dat | while read file
        pkg=$(apt-file search -x "^${file}$" | cut -d: -f1)
        [ -n "$pkg" ] && echo "$pkg" >> /tmp/packages.dat

# Remove any duplicates
$ sort -u /tmp/packages.dat > /tmp/packages.sorted

# Re-install
$ sudo apt-get -o Dpkg::Options::="--force-confmiss" install \
    --reinstall $(cat /tmp/packages.sorted)

15   Testing

Before embarking on rewriting your systems job configuration files, think very, very carefully.

We would advise strongly that before you make your production server unbootable that you consider the following advice:

  1. Version control any job configuration files you intend to change.

    You could employ the version stanza to help in this regard.

  2. Test your changes in a Virtual Machine.

  3. Test your changes on a number of non-critical systems.

  4. Backup all your job configuration files to both:

    • An alternate location on the local system

      (Allowing them to be recovered quickly if required).

    • At least one other suitable alternate backup location.

16   Daemon Behaviour

Upstart manages the running of jobs. Most of these jobs are so-called "daemons", or programs that:

To manage such daemons, Upstart expects a daemon to adhere to the following rules:

The following are recommendations if you are writing a new daemon:

17   Precepts for Creating a Job Configuration File

17.1   Determining the value of expect

The Expect section explains how to determine the value of the expect stanza. Note that you should not introduce the respawn stanza until you are fully satisfied you have specified the expect stanza correctly.

17.3   Services

  • If your job is a service, identify the correct value for the expect stanza.

    Once you have decided on the correct value:

    1. start the job:

      $ sudo start myjob
    2. Check the PID of the job matches the expected PID:

      $ actual_pid=$(pidof myapp)
      $ upstart_pid=$(status myjob | awk '{print $NF}')
      $ [ "$actual_pid" = "$upstart_pid" ] || echo "ERROR: pid "
    3. Stop the job:

      $ sudo stop myjob
    4. Ensure the PID no longer exists:

      $ [ -z "$(pidof myapp)" ] || echo "ERROR: myapp still running"
  • Only once you have specified the correct expect stanza should you introduce the respawn stanza since if you introduce it at the outset, this will just confuse your understanding, particularly if the expect stanza has been misspecified.

17.4   Ubuntu Rules ()

On Ubuntu, the following rules should be adhered to:

17.4.1   Console attributes

Jobs that specify console output or console owner should NOT modify the attributes of the console (/dev/console), for example by using tcsetattr(3).

The reason for this being that Plymouth, the graphical boot splash application, needs full control over the console on boot and shutdown.

18   Debugging

18.1   Obtaining a List of Events

To obtain a list of events that have been generated by your system, do one of the following:

18.1.1   Add --verbose or --debug to the kernel command-line

By adding --verbose or --debug to the kernel command-line, you inform Upstart to enter either verbose or debug mode. In these modes, Upstart generates extra messages which can be viewed in the system log. See initctl log-priority.

Assuming an standard Ubuntu Natty system, you could view the output like this:

grep init: /var/log/syslog

Note that until Upstart 1.3 it was difficult to get a complete log of events for the simple reason that when Upstart starts, there is no system logger running to record messages from Upstart (since Upstart hasn't started it yet!) However, Upstart 1.3 writes these "early messages" to the kernel ring buffer (see dmesg(1)) such that by considering the kernel log and the system log, you can obtain a complete list of events from the initial "startup". So, for a standard Ubuntu Oneiric system, you would do:

grep init: /var/log/kern.log /var/log/syslog

The mechanism for adding say the --debug option to the kernel command-line is as follows:

  1. Hold down SHIFT key before the splash screen appears (this will then display the grub menu).
  2. Type, "e" to edit the default kernel command-line.
  3. Use the arrow keys to go to the end of the line which starts "linux /boot/vmlinuz ...".
  4. Press the END key (or use arrows) to go to end of the line.
  5. Add a space followed by "--debug" (note the two dashes).
  6. Press CONTROL+x to boot with this modified kernel command line.

18.1.2   Change the log-priority

If you want to see event messages or debug messages "post boot", change the log priority to debug or verbose. See initctl log-priority.

18.2   See the Environment a Job Runs In

To get a log of the environment variables set when Upstart ran a job you can add simple debug to the appropriate script section. For example:

  echo "DEBUG: `set`" >> /tmp/myjob.log

  # rest of script follows...
end script

Alternatively you could always have the script log to the system log:

  logger -t "$0" "DEBUG: `set`"

  # rest of script follows...
end script

Or, have it pop up a GUI window for you:

env DISPLAY=:0.0

  env | zenity --title="got event $UPSTART_EVENTS" --text-info &
end script

For the full details, install the procenv(1) utility and run this as a job. On a Debian Sid or Ubuntu Raring (or newer) system:

$ sudo apt-get -y install procenv
$ cat <<EOT | sudo tee /etc/init/procenv.conf
exec /usr/bin/procenv
$ sudo start procenv
$ sudo cat /var/log/upstart/procenv.log

18.3   Checking How a Service Might React When Run as a Job

Before you even put your service into a Job Configuration File, try the following test which simulates an Upstart-like environment.

Assuming your service is /usr/bin/mydaemon and you want to run it as user root:

$ user=root
$ cmd=/usr/bin/mydaemon
$ su -c 'nohup env -i $cmd </dev/null >/dev/null 2>&1 &' $user

That command will run /usr/bin/mydaemon:

  • as user $user (root here, but maybe not for you if you've used setuid)
  • with no associated terminal
  • parented to init
  • with no environment

Or, if you want to set a user and a group, use sudo(8) (or maybe su(1) and newgrp(1)):

$ user=user1
$ group=group2
$ cmd=/usr/bin/mydaemon
$ ( sudo -u $user -g $group nohup env -i $cmd < /dev/null > /dev/null 2>&1 ) &

For the sudo example, you should first check that $user is able to run $cmd.

If your service is unable to run in one of these environments, it is also likely to fail when run as a Job.

18.3.1   Determining why your Service Fails to Start

You may find that your service runs fine when executed from the command-line, but does not work initially when you start testing it with Upstart. This is because the environment the service is run in when started by Upstart is potentially radically different to your interactive user (or even root user) environment.

To discover exactly what sort of environment Upstart provides, see the procenv example in See the Environment a Job Runs In.

You can also use procenv and diff(1) to determine relatively quickly how the two environments differ:

  1. Run your application or daemon (let's call it "mycmd") from the command-line (where it is expected to work) using procenv to log the environment:

    $ procenv --file=/tmp/procenv-cmdline.log --exec -- mycmd --arg1 --foo=bar
  2. Run your application from within an Upstart Job, again using procenv to log the environment:

    exec procenv --file=/tmp/procenv-job.log --exec -- mycmd --arg1 --foo=bar

    Or from a script section:

    exec procenv --file=/tmp/procenv-job.log --exec -- mycmd --arg1 --foo=bar
    end script
  3. Compare the two environments:

    $ diff /tmp/procenv-cmdline.log /tmp/procenv-job.log

18.4   Obtaining a log of a Script Section

18.4.1   Upstart 1.4 (and above)

Upstart 1.4 provides automatic logging of all job output.

See console log for further details.

18.4.2   Versions of Upstart older than 1.4

This technique relies on a trick relating to the early boot process on an Ubuntu system. On the first line below script stanza, add:

exec >>/dev/.initramfs/myjob.log 2>&1
set -x

This will ensure that /bin/sh will log its progress to the file named /dev/.initramfs/myjob.log.

The location of this file is special in that /dev/.initramfs/ will be available early on in the boot sequence (before the root filesystem has been mounted read-write).

Note that newer releases of Ubuntu mount /run/ read-writeable very early on in the boot process too.

18.5   Log Script Section Output to Syslog

There are two techniques you can use to do this:

Use the same technique as shown in Obtaining a log of a Script Section, but change the file to /dev/kmsg. This will send the data to the kernels ring buffer. Once the syslog(3) daemon starts, this data will be redirected to the system log file:

  exec >/dev/kmsg 2>&1
  echo "this data will be sent to the system log"
end script

18.7   Check a Script Section for Errors

Upstart runs your job using /bin/sh -e for safety reasons: scripts running as the root user need to be well-written! But how can you check to ensure that your script sections contain valid (syntactically correct at least) shell fragments? Simply run the init-checkconf script, which performs these checks automatically.

18.7.1   Older versions of Upstart

To check that you haven't made a (shell) syntax error in your script section, you can use sed like this:

$ /bin/sh -n <(sed -n '/^script/,/^end script/p' myjob.conf)

Or for a pre-start script section:

$ /bin/sh -n <(sed -n '/^pre-start script/,/^end script/p' myjob.conf)

No output indicates no syntax errors.

Alternatively, you could wrap this into a script like this:


[ $# -ne 1 ] && { echo "ERROR: usage: $0 <conf_file>"; exit 1; }

[ ! -f "$file" ] && { echo "ERROR: file $file does not exist" >&2; exit 1; }

for v in pre-start post-start script pre-stop post-stop
  if egrep -q "\<${v}\>" $file
    sed -n "/^ *${v}/,/^ *end script/p" $file | \
      sh -n || echo "ERROR in $v section"

And run it like this to check all possible script sections for errors:

$ myjob.conf

18.8   Debugging a Script Which Appears to be Behaving Oddly

If a script section appears to be behaving in an odd fashion, the chances are that one of the commands is failing. Remember that Upstart runs every script section using /bin/sh -e. This means that if any simple command fails, the shell will exit. For example, if file /etc/does-not-exist.cfg does not exist in the example below the script will exit before the shell runs the if test:

  grep foo /etc/does-not-exist.cfg >/dev/null 2>&1
  if [ $? -eq 0 ]
    echo ok
    echo bad
end script

In other words, you will get no output from this script if the file grep is attempting to operate on does not exist.

The common idiom to handle possible errors of this type is to convert the simple expression into an expression guaranteed to return true:

  # ensure this statement always evaluates to true
  command-that-might-fail || true

  # ditto
  another-command || :
end script

See man sh for further details.

19   Recovery

If you do something really bad or if for some reason Upstart fails, you might need to boot to recovery mode and revert your job configuration file changes. In Ubuntu, you can therefore either:

19.1   Boot into Recovery Mode

Select the "recovery" option in the Grub boot menu

This assumes that Upstart (init(8) itself) is usable.

Note that you need to hold down the SHIFT key to see the Grub boot menu.

19.2   Boot to a shell directly

If Upstart (init(8)) itself has broken, you'll need to follow the steps below. By specifying an alternate "initial process" (here a shell) it is possible to repair the system.

  1. Hold down SHIFT key before the splash screen appears (this will then display the grub menu).

  2. Type, "e" to edit the default kernel command-line.

  3. Use the arrow keys to go to the end of the line which starts "linux /boot/vmlinuz ...".

  4. Press the END key (or use arrows) to go to end of the line.

  5. Add a space followed by "init=/bin/sh".

  6. If the line you are editing contains "quiet" and/or "splash", remove them.

  7. Press CONTROL+x to boot with this modified kernel command line.

  8. When the shell appears you will need to remount the root filesystem read-write like this:

    # mount -oremount,rw /

You can now make changes to your system as necessary.

20   Advanced Topics

20.1   Changing the Default Shell

By default, Upstart uses "/bin/sh" to execute script sections. If you wish to change this behaviour, you have the following options:

  • Link /bin/sh to your chosen shell [13].

  • Copy your chosen shell to /bin/sh.

  • Recompile Upstart specifying an alternative shell as follows:

    # XXX: Note the careful quoting to retain double-quotes around the shell!
    export CFLAGS=-DSHELL='\"/bin/bash\"'
    ./configure && make && sudo make install

    Note that you should consider such a change carefully since Upstart has to rely upon the shell. Remember too that Upstart runs all script sections as the root user.

  • Use a "here document" (assuming your chosen shell supports them) within the Job Configuration Files you wish to run with a different shell:

    /bin/bash <<EOT
    echo "Hi - I am running under the bash shell"
    echo "and so am I :)"
    end script

    Note that currently, this technique is the only way (without modifying the Upstart source code) to run a shell without specifying the "-e" option (see dash(1) or bash(1) for details).

20.2   Running a script Section with Python

To run a script section with Python:


python - <<END

from datetime import datetime

today ="%A")

fh = open("/tmp/file.txt", "w")
print >>fh, "Today is %s" % today


end script

20.3   Running a script Section with Perl

To run a script section with Perl:


perl - <<END

use strict;
use warnings;
use POSIX;

my $fh;
my $today = POSIX::strftime("%A", localtime);

open($fh, ">/tmp/file.txt");
printf $fh "Today is %s\n", $today;


end script

21   Development and Testing

21.1   Warnings

  • Upstart runs as root so has full system privileges.

  • If Upstart crashes...:

    Kernel panic - not syncing: Attempted to kill init! exitcode=0x00000100
    [ 2.745566]
    [ 2.751931] Pid: 1, comm: false Not tainted 3.5.0-15-generic #22-Ubuntu
    [ 2.755489] Call Trace:
    [ 2.757068]  [<c15be842>] panic+0x81/0x17b
    [ 2.759206]  [<c104a6a5>] do_exit+0x745/0x7a0
    [ 2.761602]  [<c104a9a4>] do_group_exit+0x34/0xa0
    [ 2.764162]  [<c104aa28>] sys_exit_group+0x18/0x20
    [ 2.765231]  [<c15c8a94>] syscall_call+0x7/0xb

    ... your kernel panics!

  • Unlike the kernel, if a new version of Upstart fails to work at all, there is no easy fix.

21.2   Precautions and Practises

Precautions and Practises:

  • Every function asserts its arguments. This allows simple programming bugs to be found very quickly ("fail fast").
  • Every function that returns a value must be checked and exceptions handled. GCC helps in this respect with the __attribute__ ((warn_unused_result)) function attribute.
  • Every function that returns newly-allocated memory has its prototype decorated using __attribute__ ((malloc)).
  • No compiler warnings are allowed (-Wall -Werror).
  • Every function or logical unit of functionality must have an associated set of tests.
  • Every build of Upstart must pass all NIH and Upstart tests before being made available to users.
  • The code is very well tested(using physical and virtual hardware, all architectures and containers).
  • All code is peer-reviewed.
  • All changes to the main lp:upstart code branch in Launchpad now automatically generate a mail to the Upstart mailing list.
  • All bzr merge proposals raised on Upstart also result in a mail to the Upstart mailing list.
  • Where possible, all new features add a --no--<feature> command-line option (allowing the feature to be disabled to provide a fall-back mechanism).

21.3   Code Style

  • Use tabs.
  • Every function, macro, structure, typedef and variable must be documented.
  • Every function must specify what is returned on success and failure.
  • Every function must check all possible parameters using .

See file: upstart:HACKING

21.4   Development Advice

  • KISS and KIRS ("keep it readable silly")

    "Clever" code often outwits the author.

    Prefer to keep it simple, elegant and most of all readable. Bit-twiddlers and IOCCC champions need not apply.

  • Do not use system calls or library calls if NIH already provides an alternative. That means:

    • No malloc(), calloc(), strtok(), sprintf(), et cetera.
    • You really need to familiarise yourself with NIH by reading the NIH source and the Upstart source.
  • Don't just read the NIH and Upstart source, read the test code - it has comments too! ;-)

  • Write code to be testable.

  • Always consider security and performance.

    • DoS possibility?
    • Get the code security-reviewed.
  • If you plan to work on some huge feature that will take you 6 months of effort, PLEASE alert the developers via the mailing list BEFORE you start since:

    • We may already be working on such a feature.
    • If your design doesn't fit in with the project, you're potentially facing a lot of avoidable rework.
  • Always test on a range of hardware:

    • Physical and virtual.
    • 32-bit and 64-bit.
    • Intel/ARM/etc.

21.5   Setting up an Upstart Development Environment

$ sudo apt-get install build-dep upstart # cheat :)
$ bzr branch lp:upstart
$ cd upstart
$ ./configure --disable-silent-rules --enable-compiler-warnings \
    --disable-compiler-optimisations --disable-linker-optimisations \
$ export CFLAGS="-fstack-protector --param=ssp-buffer-size=4 -Wformat -Werror=format-security"
$ make
$ cscope -Rbq && ctags
  • You need to build the code before indexing since D-Bus bindings are auto-generated (using nih-dbus-tool).
  • You need to use all those flags to enable all the compiler checks.

21.6   Setting up an Upstart+NIH Development Environment

Since Upstart makes such heavy use of NIH, it is often useful to build both Upstart and link it to a debug symbols build of NIH:

$ sudo apt-get install build-dep upstart libnih1 # cheat :)
$ prefix=/testing
$ mkdir $prefix
$ export PKG_CONFIG_PATH=${prefix}/lib/pkgconfig:$PKG_CONFIG_PATH
$ export ACDIR=${prefix}/share/aclocal:$ACDIR
$ export CFLAGS="-fstack-protector --param=ssp-buffer-size=4 -Wformat \
                 -Werror=format-security -ggdb3 -fno-inline"
$ bzr branch lp:libnih
$ cd libnih
$ ./configure --disable-silent-rules --enable-compiler-warnings \
    --disable-compiler-optimisations --disable-linker-optimisations \
    --enable-compiler-coverage && make && make install
$ cd -
$ bzr branch lp:upstart
$ cd upstart
$ ./configure --disable-silent-rules --enable-compiler-warnings \
    --disable-compiler-optimisations --disable-linker-optimisations \
    --enable-compiler-coverage && make && make install

21.7   Upstart Objects

  • Event represents an event.

  • ConfSource represents a type of configuration source (file or directory) and

    includes inotify watches.

  • ConfFile represents the jobs ".conf" file name, but also

    has a pointer to its contents (see below).

  • JobClass represents the jobs ".conf" file contents.

  • Job represents a running instance of job.

  • Session represents a user session for user jobs, or a chroot.

  • Log represents job log data (data that a single job process has produced on its standard output and standard error).

  • Blocked is used to handle hook and method events types.

    (See hooks and methods).

See upstart-objects-diagram.

21.8   Unit Tests

Every major feature in Upstart needs to be accompanied with comprehensive unit tests. To run the tests:

$ autoreconf -fi
$ ./configure --enable-compiler-coverage ...
$ make check 2>&1|tee make-check.log

Note that as of Upstart 1.3, some of these tests cannot be run from within a chroot(2) environment unless D-Bus is installed and configured within the chroot. This scenario is detected, a warning about bug 728988 is logged and those tests are automatically skipped. Hence, to run all the tests, please ensure you run "make check" outside of a chroot(2) environment.

21.8.1   Building Within a Chroot

Some of the unit tests assume a full environment, including a controlling terminal. If you wish to build an Upstart package on a Debian or Ubuntu system, note that although the pbuilder(8) tool will work as expected, currently sbuild(1) does not provide a controlling terminal which causes tests to fail. See [23] and [24].

21.8.2   Statistics

At the time of writing, the number of Upstart tests, and tests for the NIH Utility Library used by Upstart are:

Unit Test Statistics.
Application Test Count
Upstart unit tests 1368
NIH Utility Library 2863
Total 4231

importance of the test-suite cannot be overstated: it's one of the main "safety-nets" to ensure the behaviour of NIH and Upstart is assured.

To run the test suite for NIH or Upstart, simply run the following as a non-privileged user:

make check

21.8.3   Test Coverage

To check the test coverage after running the tests, look at each file using gcov(1):

$ cd init
$ gcov -bf event.c

21.9   Enable Full Compiler Warnings

If you want to start submitting changes to Upstart, you need to ensure you build it as follows to catch any warnings and errors the compiler can flag:

./configure --disable-silent-rules --enable-compiler-warnings --disable-compiler-optimisations --disable-linker-optimisations --enable-compiler-coverage

21.10   Running Upstart as a Non-Privileged User

Upstart 1.3 introduced a number of options to help with testing. The "--session" command-line option allows you to run Upstart as a non-privileged user since it makes Upstart connect to the D-Bus session bus for which each user has their own:

$ /sbin/init --session --debug --confdir $HOME/conf/ --no-sessions

This is useful since you can now try out new features, debug with GDB, et cetera without having to install Upstart and run it as root. Once you've got your second instance of Upstart running, you can then use the same option on initctl to manipulate jobs:

$ initctl --session emit foo

The caveat here is that running Upstart as a non-privileged user with a PID other than 1 changes its behaviour slightly. So, only use this technique for unit/functional testing and remember that any changes you post for inclusion should have been tested in a real scenario where Upstart is run as root and used to boot a system.

21.11   Useful tools for Debugging with D-Bus

If you are debugging initctl(8), you'll need to understand D-Bus. These tools are invaluable:

21.12   Debugging a Job

There is a magic stanza called debug which will start the job via fork(2) and then pause it. This can be useful. Assuming you have a job "debug.conf" such as:

# XXX: magic stanza!

end script

You could now trace the job process like this:

# start debug
debug start/running, process 12345
# strace -p 12345 -o /tmp/debug.log -Ff -s 1024 -v
status debug debug stop/waiting

After the call to start, the job process will be "running", but paused. The strace(1) will resume the job and you will then have a log of what happened in file "/tmp/debug.log".

21.13   Debugging Another Instance of Upstart Running as root with PID 1

21.13.1   Method 1 (crazy)

Caveat Emptor: this is somewhat crazy, but if you really want to do this:

$ sudo \
  gdb --args \
  init/init --debug --confdir /my/conf/dir --no-startup-event

This uses the Clone tool, which is very similar to unshare(1) but allows you to put a process into a new PID namespace.

21.13.2   Method 2 (saner)

Use a container technology such as LXC, that simplifies the access to namespaces. For example [12]:

$ sudo lxc-start -n natty
$ upstart_pid=$(pgrep -f /sbin/init|grep -v '^1$')
$ sudo gdb /sbin/init $upstart_pid

Like the example above, here we use gdb to debug Upstart running as root with PID 1, but with thanks to LXC, the container is fully isolated from the host system using namespaces. See lxc(7) for details of LXC on Ubuntu.

21.14   NIH

Grab the code from the NIH Utility Library page.

The NIH documentation is with the code:

  • Header files provide introductory details.
  • Every function, macro and variable is documented immediately above it.

References in the sections below give locations of file in the NIH source.

21.14.1   Memory Handling

Do not use malloc(), calloc(), realloc() or free() when working with Upstart. Rely instead on the NIH memory routines:

  • Low-level memory allocation is handled using nih_alloc() and nih_realloc().

  • It is more normal to use nih_new(parent, type) though.

  • To free memory, use nih_free():

    typedef struct foo {
        int i;
    } Foo;
    Foo *foo = nih_new (NULL, Foo);
    foo->i = 123;
        /* time passes... */
    nih_free (foo);


NEVER free memory using nih_free() that NIH did not allocate!

See: nih/alloc.[ch]

Like C++, NIH can perform automatic cleanup when objects go out of scope. The most magical part of NIH is nih_local.

Question: is the following code leaking memory?

void foo (void)
    nih_local char *string = nih_strdup (NULL, "hello, world");

    nih_message ("%s", string);

Answer: No!

  • nih_local is syntactic sugar to tell the compiler that the memory that the variable it applies to ("string") should be freed when the last reference to it is dropped. This happens when the variable goes out of scope at the end of foo().


ALWAYS assign nih_local variables to NULL to avoid memory corruption issues if the variable is not assigned for some code path!

21.14.2   The NIH Parent Pointer

Most NIH routines take a void *parent as their first parameter.

This parent pointer can be NULL as shown below:

nih_strdup (NULL, "hello, world");

If the parent is not NULL, NIH will automatically add an appropriate reference such that when the parent is freed, so are its child objects.

Consider this example:

void bar (void)
    typedef struct thing {
        char *str;
    } Thing;

    nih_local Thing *thing = nih_new (NULL, Thing);

    /* XXX: note that we specify the parent as 'thing' */
    thing->str = nih_strdup (thing, "first string");

Two memory allocations have been performed:

  • thing
  • thing->str

And yet when bar() exits, there is no leak because NIH knows that thing->str is a "child" of thing and will do-the-right-thing (TM) and free both chunks of memory!

Here is another subtle example:

void bar (void)
    typedef struct thing {
        char *str;
    } Thing;

    nih_local Thing *thing = nih_new (NULL, Thing);

    /* XXX: note that we specify the parent as 'thing' */
    thing->str = nih_strdup (thing, "a string value");

    /* now, let's reassign the pointer */
    thing->str = nih_strdup (thing, "another string value");

Surely, there must be a leak now since we've re-assigned thing->str?

In fact, there is no leak because both the strings that we've assigned to thing->str have specified the same parent: thing. So, the reference to a string value has not been lost and both string values will be freed correctly when thing goes out of scope!

21.14.3   nih_free()

However, sometimes using nih_local is not appropriate. In the example below, we manually free the memory using nih_free():

void bar (void)
    typedef struct thing {
        char *str;
    } Thing;

    Thing *thing = nih_new (NULL, Thing);

    /* XXX: note that we specify the parent as 'foo' */
    thing->str = nih_strdup (thing, "first string");

    /* "manually" free thing _and_ thing->str */
    nih_free (thing);

Here we use nih_free() to force NIH to free up memory.


NEVER call nih_free() on an nih_local variable!

21.14.4   NIH_MUST()

The example so far have not checked for error conditions. Here's how we could handle an out-of-memory scenario:

nih_local char *string = NULL;

string = nih_strdup (thing, "first string");

if (! string) {
  /* handle the error */

However, this tends to lead to code littered with error checking. There is a common NIH idiom that avoids such problems:

nih_local char *string = NULL;

string = NIH_MUST (nih_strdup (thing, "first string"));

  /* string is now guaranteed to have the expected error */

See file: nih/macros.h

NIH_MUST() will evaluate its argument until it returns a value.


NIH_MUST() will try forever to grab the memory required.

That could lead to Upstart going into a tight loop and effectively killing your machine.

However, realistically, Upstart only ever allocates small chunks of memory and if /sbin/init, running as root is unable to allocate a few bytes of memory, you machine has big problems.

21.14.5   Error Handling

If a function detects a failure, it must return a suitable error value. However, it may be appropriate to raise an exception. You'll know if a function raises an exception since it will be documented like this:

Returns: zero on success, negative value on raised error.

A "raised error" refers to an NihError object being raised when the function detects an error.

Therefore, it is the callers responsibility to:

  • Check the return code of every function that returns a value.
  • Handle raised errors appropriately (and immediately!)

See: nih/error.[ch]

Let's look at an example:

char *
num_to_str (int i)
    if (i % 2)
        return NIH_MUST (nih_sprintf (NULL, "%d", i));

    nih_error_raise_no_memory ();
    return NULL;

main (int argc, char *argv[])
    nih_local char *s = NULL;

    s = num_to_str (1);
    nih_message ("got: '%s'", s);

    /* force error scenario */
    s = num_to_str (2);
    if (! s) {
        /* retrieve the error */
        err = nih_error_get ();

        /* display it */
        nih_message ("%s:%d:%s:%d:%s",
        /* clear the error */
        nih_free (err);

    exit (EXIT_SUCCESS);
}   Impact of Ignoring a Raised Error

An example of code that ignores a raised error:

char *
num_to_str (int i)
    if (i % 2)
        return NIH_MUST (nih_sprintf (NULL, "%d", i));

    nih_error_raise_no_memory ();
    return NULL;

main (int argc, char *argv[])
    nih_local char *s = NULL;

    /* ok */
    s = num_to_str (1);
    nih_message ("got: '%s'", s);

    /* force error scenario */
    s = num_to_str (2);

    /* Oops - forgot to check return! */
    nih_message ("got: '%s'", s);

    exit (EXIT_SUCCESS);


got: '1'
got: '(null)'
(null):test_nih_error.c:38: Unhandled error from num_to_str: Cannot
allocate memory
[1]    20476 abort (core dumped)  bin/test_nih_error

The reason this crashes is that NIH installs an atexit(3) handler which checks for any NihError errors that have not been handled on exit.

Of course, in the case of Upstart, it never exits so failing to handle an error will result in an assertion failure the next time an error object is raised.

  • To raise an exception when ERRNO gets set, use:
    • nih_error_raise_system()
    • nih_return_system_error()
  • To raise an arbitrary exception, use:
    • nih_error_raise(number, message)
    • nih_error_raise_printf(number, format, ...)
    • nih_return_error(retval)

See file: nih/error.h

21.14.6   Output

NIH has a rich set of output routines:

  • nih_debug()
  • nih_info()
  • nih_message()
  • nih_warn()
  • nih_fatal()
  • nih_fatal()

All routines take a format string and arguments like printf(3):

int   i = 123;
char *s = "hello, world";
nih_debug ("s='%s', i=%d", s, i);

Like syslog(3), NIH will only display message made with the above calls if the log priority is appropriate.

To change the priority, use --verbose, --debug, or programatically call nih_set_priority().

By default, output goes to standard output, but early in its initialisation, it redirects output to the kernel ring buffer using:

nih_log_set_logger (logger_kmsg);

See file: nih/logging.[ch]

21.15   Creating a New Object

21.15.1   Template for a new "foo"

 * foo:
 * @entry: list header,
 * @name: name of foo,
 * @value: value of foo.
 * Structure to hold a foo.
 * << XXX: more details here >>.
 typedef struct foo {
     NihList   entry;
     char     *name;
     int       value;
 } Foo;

  * foos:
  * List of all foos. << XXX: more details here >>
 NihList *foos;

  * Initilise the foos list.
 void foo_init (void)
     if (! foos)
         foos = NIH_MUST (nih_list_new (NULL));

 Foo * foo_new (void *parent, const char *name, int value)
         __attribute__ ((warn_unused_result, malloc));

  * foo_new:
  * @parent: parent of new foo,
  * @name: name of foo,
  * @value: value of foo.
  * Returns: Newly allocated foo, or NULL on insufficient memory.
  Foo *
  foo_new (void *parent, const char *name, int value)
      Foo *foo;

      assert (name); /* check all args possible */
      foo_init (); /* initialise the subsystem */

      /* create the object */
      foo = NIH_MUST (nih_new (parent, Foo));

      /* initialise the embedded list */
      nih_list_init (&foo->entry);

      /* save values */
      foo->name = NIH_MUST (nih_strdup (foo, name));
      foo->value = value;

      /* Add object to list of known foos */
      nih_list_add (foos, &source->entry);

      /* explain how objects should be disposed of */
      nih_alloc_set_destructor (foo, nih_list_destroy);

      return foo;

      nih_free (foo);
      return NULL;

21.15.2   Basic Test Example for a New "foo"

Foo  *foo;
char *str;

TEST_FEATURE ("with parent");

foo_init ();


str = nih_strdup (NULL, "hello");
TEST_NE_P (str, NULL);

foo = foo_new (str, "foo", 123);
TEST_NE_P (foo, NULL);

TEST_ALLOC_SIZE (foo, sizeof (Foo));

TEST_FREE_TAG (foo->name);

TEST_EQ (foo->value, 123);
TEST_EQ_STR (foo->name, "foo");
TEST_ALLOC_PARENT (foo->name, foo);

nih_free (foo);
TEST_FREE (foo->name);

nih_free (str);

21.16   Adding a new initctl command

21.16.1   Adding a New non-Job Command

  1. Add a new function called "<name>_action()" to util/initctl.c where "<name>" is the name of the new command the user will type on the command-line ("initctl <name>") with all hyphens ("-") converted to underscores ("_").

    Example: "reload_configuration_action()" for the "reload-configuration" command-line command.

  2. Make "<name>_action()" call "upstart_<name>_sync()", which will be an auto-generated function (see below).

    Example: "reload_configuration_action()" calls "upstart_reload_configuration_sync()".

  3. Add a new D-Bus method corresponding to "<name>" in "camel-case" to:


    Example: Add the following for the "reload-configuration" command:

    <method name="ReloadConfiguration">
  4. Add implementation to "init/control.c" as "control_<name>()".

    Example: add "control_reload_configuration()".

21.16.2   Adding a New Job Class Command

Process is as per Adding a new non-Job Command, but rather than modifying file "dbus/com.ubuntu.Upstart.xml", you must modify file:


... and then add a function to "init/job_class.c".

21.16.3   Adding a New Job Command

Process is as per Adding a new non-Job Command, but rather than modifying file "dbus/com.ubuntu.Upstart.xml", you must modify file:


... and then add a function to "init/job.c".

21.16.4   Generating the D-Bus Bindings

After following the steps above to add a new initctl command, run "make" and observe the the nih-dbus-tool utility gets calls to convert your XML definitions into auto-generated code:

/usr/bin/nih-dbus-tool \
                --package=upstart \
                --mode=object --prefix=control \
                --default-interface=com.ubuntu.Upstart0_6 \
/usr/bin/nih-dbus-tool \
                --package=upstart \
                --mode=object --prefix=job_class \
                --default-interface=com.ubuntu.Upstart0_6.Job \
/usr/bin/nih-dbus-tool \
                --package=upstart \
                --mode=object --prefix=job \
                --default-interface=com.ubuntu.Upstart0_6.Instance \


NIH provides a rather clever macro called TEST_ALLOC_FAILED; it accepts a code block and will execute that block 1 + N times where N is the number of NIH memory allocation calls made within the block.

  • The first time through, the macro counts the number of NIH allocation calls.
  • Each subsequent time through, it causes the Nth call to an NIH memory allocation routine to fail.

This exercises fully for example a function which returns a newly-allocated object (and which may make any number of calls to the NIH memory allocation routines).

Essentially, it ensures your handling of memory allocation failures are correct.

21.17.1   Improved Test Example for a New "foo" (with a bug)

We can now modify our previous example to also use TEST_ALLOC_FAIL. Note that this version contains a bug! Can you spot it?:

Foo  *foo;
char *str;

TEST_FEATURE ("put text here");

foo_init ();

    TEST_LIST_EMPTY (foos);

    str = nih_strdup (NULL, "hello");
    TEST_NE_P (str, NULL);

    foo = foo_new (str, "foo", 123);
    if (test_alloc_failed) {
        TEST_EQ_P (foo, NULL);


    TEST_ALLOC_SIZE (foo, sizeof (Foo));
    TEST_EQ (foo->value, 123);
    TEST_EQ_STR (foo->name, "foo");

    nih_free (str);


If you need to guarantee that particular memory allocations within the do not fail, wrap those in a call to TEST_ALLOC_SAFE:

        /* Memory allocations will work here */

    /* Memory allocations will be sequentially FAILED here */

21.18.1   Final Test Example for a New "foo"

Using TEST_ALLOC_FAIL, we can now fix the example to be:

Foo  *foo;
char *str;

TEST_FEATURE ("put text here");

foo_init ();

        TEST_LIST_EMPTY (foos);

        str = nih_strdup (NULL, "hello");
        TEST_NE_P (str, NULL);

    foo = foo_new (str, "foo", 123);
    if (test_alloc_failed) {
        TEST_EQ_P (foo, NULL);


    TEST_ALLOC_SIZE (foo, sizeof (Foo));
    TEST_EQ (foo->value, 123);
    TEST_EQ_STR (foo->name, "foo");

    nih_free (str);

21.19   Basic Debugging

Don't underestimate the usefulness of two very simple techniques:

  • sudo strace -p 1 -fFv -s 1024
  • nih_fatal("\%s:\%d", __func__, __LINE__);

21.20   Debugging Upstart as a Non-Privileged User

With the right command-line options, it's possible to run Upstart as a normal non-privileged user:

$ make
$ mkdir /tmp/conf /tmp/log
$ cp *.conf /tmp/conf
$ gdb init/init --confdir /tmp/conf --logdir /tmp/log --no-sessions --session --debug

This is a useful technique but be aware that the behaviour of Upstart running as a non-privileged user is slightly different to running it as root with PID 1.

21.21   Debugging Upstart as root

It is in fact possible to debug /sbin/init using gdb as user root on a running system!

  • Build upstart with -ggdb3 and install to /sbin/ for example.
  • sudo gdb /sbin/ 1

21.22   Debug Tip Using Destructors

If can be useful to register a custom destructor for your object as a debug aid:

int foo_destructor(void *ignored)
    /* Do something */
    return 1;

Foo *
foo_new (void *parent)
    Foo *foo = NIH_MUST (nih_new (parent, Foo));

    /* ... */

    /* Call foo_destructor when object is destroyed */
    nih_alloc_set_destructor (foo, foo_destructor);

    return foo;

Now, whenever a Foo is freed, foo_destructor() will be called.

Note that child objects of the Foo object that foo_destructor() is being called for and the parent references and the object itself will be freed - the destructor is for very specialist operations, such as debugging.

21.22.1   Lists

Here's an example of using NIH lists:

typedef struct bar {
    NihList  entry;
    char    *str;
} Bar;

main (int argc, char *argv[])
    int i;
    nih_local NihList *args = NULL;

    args = NIH_MUST (nih_list_new (NULL));

    /* store all arguments in a list */
    for (i = 1; i < argc; ++i) {
        Bar *bar = NIH_MUST (nih_new (args, Bar));
        nih_list_init (&bar->entry);
        bar->str = NIH_MUST (nih_strdup (bar, argv[i]));
        nih_list_add (args, &bar->entry);
    i = 1;

    /* display all arguments by iterating over list */
    NIH_LIST_FOREACH (args, iter) {
        Bar *bar = (Bar *)iter;
        nih_message ("argument %d='%s'", i, bar->str);

    return (0);
  • NIH lists are designed to be embedded within some other structure.
  • Create a list dynamically using nih_list_new().
  • Initialize a static list using nih_list_init().
  • Add one list to another using nih_list_add().
  • Iterate a list using NIH_LIST_FOREACH().

See file: nih/list.[ch]   Removing Elements from a List

An example showing how to remove an element from a list:

NihList      *entry_list;
NihListEntry *entry;

entry_list = NIH_MUST (nih_list_new (NULL));

entry = NIH_MUST (nih_list_entry_new (entry_list));
entry->str = NIH_MUST (nih_strdup (entry, "hello"));
nih_list_add (entry_list, &entry->entry);

entry = NIH_MUST (nih_list_entry_new (entry_list));
entry->str = NIH_MUST (nih_strdup (entry, "world"));
nih_list_add (entry_list, &entry->entry);

entry = (NihListEntry *)nih_list_remove (entry_list);
nih_free (entry_list);

Freeing entry_list frees the "hello`" *and* the "`world`" entries since although the "``world" entry was removed from its containing list, we did NOT break the reference between that entry and its parent (entry_list).

If we had wanted to break the reference, we could have used nih_ref() and nih_unref() to:

  • Add a reference for this entry to a new parent.
  • Remove the existing reference between entry_list and the entry.

Another method for removing an entry from a list is whilst iterating it:

NIH_LIST_FOREACH_SAFE (entry_list, iter) {
    NihListEntry *entry = (NihListEntry *)iter;
    nih_free (entry);
  • Note that we are now using . Do NOT attempt to remove a list entry whilst iterating a list using .
  • It is NOT allowed to iterate a list whilst it is already being iterated. Therefore, you need to be very careful that your function is not being called from within a foreach-loop.   Moving an Element Between Lists

An example showing moving an element from one list to another:

NihList      *list1;
NihList      *list2;
NihListEntry *entry;

list1 = NIH_MUST (nih_list_new (NULL));
list2 = NIH_MUST (nih_list_new (NULL));

/* Create entry and add to list1 */
entry = NIH_MUST (nih_list_entry_new (list1));
nih_list_add (list1, &entry->entry);

/* Fully move entry to list2 */
nih_list_add (list2, &entry->entry);
nih_ref (entry, list2);
nih_unref (entry, list1);

/* Frees list1, but not entry */
nih_free (list1);

/* Frees list2 AND entry */
nih_free (list2);

21.22.2   Hashes

NIH Hashes are actually "hashed lists" (essentially arrays of lists):

NihHash *
nih_hash_new (const void      *parent,
              size_t           entries,
              NihKeyFunction   key_function,
              NihHashFunction  hash_function,
              NihCmpFunction   cmp_function);

However, the more common way to create a hash is via:

typedef struct foo {
    NihList  entry;
    char    *name;
} Foo;

 * foos:
 * List of all foos. << XXX: more details here >>
NihHash *foos;

 * Initilise the foos hash.
void foo_init (void)
    if (! foos)
        foos = NIH_MUST (nih_hash_string_new (NULL, 0));

Foo *
foo_new (void *parent, const char *name)
    Foo *foo;

    assert (name);
    foo_init (); /* initialise the subsystem */

    /* create the object */
    foo = NIH_MUST (nih_new (parent, Foo));

    /* initialise the embedded _list_ */
    nih_list_init (&foo->entry);

    nih_hash_add (foos, &foo->entry);

    return foo;
}   Using Hashes

To iterate a hash, use NIH_HASH_FOREACH():

NIH_HASH_FOREACH (foos, iter) {
    Foo *foo = (Foo *)iter;

    /* do something with foo */

To find an entry in a hash, use nih_hash_lookup():

Foo *foo;

foo = (Foo *)nih_hash_lookup (foos, "hello");
if (foo) {
    /* ... */

Alternatively, if there are multiple entries for a particular "hash bucket", use nih_hash_search().

See: nih/hash.[ch]   nih_hash_string_new()

nih_hash_string_new()``is "magic" *BUT* to use it *the first structure element **after** the element **must** be a "``char *" that will uniquely represent that hash entry*.

If a simple string is not sufficient for your purposes, you will need to use nih_hash_new() and will also have to specify the NihKeyFunction, NihHashFunction and NihCmpFunction.

Analogous to NIH_LIST_FOREACH_SAFE, there is also a NIH_HASH_FOREACH_SAFE facility for removing hash entries whilst iterating the hash.

21.22.3   Trees

A basic example of NIH trees:

typedef struct foo {
    NihTree node;
    int     value;
} Foo;

NihTree *tree;
Foo     *foo;

tree = NIH_MUST (nih_tree_new (NULL));
foo = NIH_MUST (nih_new (tree, Foo));
nih_tree_init (&foo->node);
foo->value = 123;

nih_tree_add (tree, &foo->entry, NIH_TREE_LEFT);

To iterate a tree:

  • NIH_TREE_FOREACH() (in-order traversal)
  • NIH_TREE_FOREACH_PRE() (pre-order traversal)
  • NIH_TREE_FOREACH_POST() (post-order traversal)

See: nih/tree.[ch]

Example of iterating a tree using in-order traversal:

NIH_TREE_FOREACH (tree, iter) {
    Foo *foo = (Foo *)iter;
    /* ... */

21.22.4   Avoiding Problems

What's wrong with this code?:

/* XXX: this code is incorrect! */
void foo (const char *string)
    nih_local char *str;
    nih_assert (string);

    if (! strcmp ("foo", string)) {
        str = NIH_MUST (nih_strdup (NULL, "bar"));
        bar (str);

The problem here is that str is not always assigned a value, so if string is not foo, the results of this function are undefined - it could result in a crash!!

The example below contains two memory leaks:

NihList      *entry_list;
NihListEntry *entry;

entry_list = NIH_MUST (nih_list_new (NULL));

entry = NIH_MUST (nih_list_entry_new (NULL));
entry->str = NIH_MUST (nih_strdup (NULL, "hello"));
nih_list_add (entry_list, &entry->entry);

nih_free (entry_list);
  • entry is not freed. To resolve, either:
    • Make its parent pointer non-NULL (recommended).
    • Call nih_free().
  • entry->str is not freed. To resolve, either:
    • Set its parent pointer to entry (recommended).
    • Call nih_free (entry->str).

21.23   Debugger Magic

Debugging in gdb initially seems rather difficult, but you just need to know the right tricks. The complication comes from the fact that Upstart uses the NIH Utility Library, which uses macros (such as NIH_LIST_FOREACH and NIH_HASH_FOREACH) for performance.

However, how do you access a data structure such as an NihList whose only method of iteration is a macro? Like this:

21.23.1   NihList

# first entry
(gdb) print *(JobClass *)job_classes->next

# 2nd entry
(gdb) print *(JobClass *)job_classes->next->next

# 3rd entry
(gdb) print *(JobClass *)job_classes->next->next->next

# ConfSource NihWatch for 1st entry in conf_sources list
(gdb) print *((ConfSource *)conf_sources->next)->watch

21.23.2   NihHash

# size of JobClass->instances hash list
# XXX: this is the capacity, *NOT* the number of entries!
print class->instances->size

# first entry in job_classes global hash
print *(JobClass *)job_classes->bins->next

21.23.3   nih_iterators

Alternatively, you can make use of the "unofficial" NIH Iterators which provide functional versions of the standard NIH macros and a few extras. Note that these are ONLY for testing and debugging!

  • nih_list_foreach():

     * nih_list_foreach:
     * @list: list,
     * @len: optional output parameter that will contain length of list,
     * @handler: optional function called for each list entry,
     * @data: optional data to pass to handler along with list entry.
     * Iterate over specified list.
     * One of @len or @handler may be NULL.
     * If @handler is NULL, list length will still be returned in @len.
     * If @handler returns 1, @len will be set to the number of list entries
     * processed successfully up to that point.
     * Returns: 0 on success, or -1 if handler returns an error.
    nih_list_foreach (const NihList *list, size_t *len, NihListHandler handler, void *data);
  • nih_hash_foreach():

     * nih_hash_foreach:
     * @hash: hash,
     * @len: optional output parameter that will contain count of hash entries,
     * @handler: optional function called for each hash entry,
     * @data: optional data to pass to handler along with hash entry.
     * Iterate over specified hash.
     * One of @len or @handler may be NULL.
     * If @handler is NULL, count of hash entries will still be returned in @len.
     * If @handler returns 1, @len will be set to the number of hash entries
     * processed successfully up to that point.
     * Returns: 0 on success, or -1 if handler returns an error.
    nih_hash_foreach (const NihHash *hash, size_t *len,
                NihListHandler handler, void *data);
  • nih_tree_foreach():

     * nih_tree_foreach:
     * @tree: tree,
     * @len: optional output parameter that will contain count of tree nodes,
     * @handler: optional function called for each tree node,
     * @data: optional data to pass to handler along with tree node.
     * Iterate over specified tree.
     * One of @len or @handler may be NULL.
     * If @handler is NULL and @len is non-NULL, count of tree nodes will
     * still be returned in @len.
     * If @handler returns 1, @len will be set to the number of tree nodes
     * processed successfully up to that point.
     * Returns: 0 on success, or -1 if handler returns an error.
    nih_tree_foreach (NihTree *tree, size_t *len,
                NihTreeFilter handler, void *data);

These routines allow us to also provide trivial implementations of the following convenience functions:

  • nih_list_count()
  • nih_hash_count()
  • nih_tree_count()

21.24   Development Utilities


The upstart-menu utility allows /sbin/init versions you wish to boot with to be selected using a friendly menu. You can also select a shell. scans /sbin/ for init version and presents a list, most recently modified version first:

upstart menu main screen

Main screen of

The utility also allows you to specify options (it automatically shows you a list of available options for the version of the program you have selected):

upstart menu options screen showing the options screen.   Enabling ,

To enable

  1. Copy file to /sbin/
  2. Make the file executable.
  3. Update /etc/default/grub such that GRUB_CMDLINE_LINUX is modified to:
    • Remove "quiet" and "splash".
    • Add "init=/sbin/".
  4. Update grub: "sudo update-grub".
  5. Reboot!

21.25   Gotchas

  • Passing NULL to nih_free(): unlike free(3), nih_free() does not allow a NULL parameter.
  • Running make check as root (tests will fail).
  • Debugging a failing memory-checking test by littering test code with calls to nih_debug()... which calls nih_alloc().
  • Forgetting to install either /sbin/init or /sbin/initctl when you modify the D-Bus interface to Upstart (if you're lucky, you'll get a crash, else very odd behaviour! :-)
  • Not checking for existing init and test_* processes still running from a previous failed test run when you run make check.

22   Known Issues

22.1   Restarting Jobs with Complex Conditions

The and and or operators allowed with start on and stop on do not work intuitively: operands to the right of either operator are only evaluated when the specified event is emitted. This can lead to jobs with complex start on or stop on conditions not behaving as expected when restarted. For example, if a job specifies the following condition:

start on A and (B or C)

When the events "A" and "B" are emitted, the condition is satisfied so the job will be run. If the job fails to start, or is stopped later, there is no guarantee that "A" will be emitted again, and the fact that it happened before is no longer known to Upstart. Meanwhile, events "C" or "B" may occur, but the job will not be transitioned back to a start goal, until event "A" is emitted again.

22.1.1   Advice

To minimise the risk of being affected by this issue, avoid using complex conditions with jobs which need to be restarted.

22.2   Using expect with script sections

Using the expect stanza with a job that uses a script section will lead to trouble if your script spawns any processes (likely!). Consider:

expect fork
  ARGS=$(cat /etc/default/grub)
  exec echo "ARGS=$ARGS" > /tmp/myjob.log
end script

This job configuration file is somewhat nonsensical, but it does demonstrate the problem. The main issue here is that by specifying expect fork, Upstart will attempt to follow only the first fork(2) call. The first process that this job will spawn is... cat(1), NOT echo. As such, starting the job will show something like this:

# start myjob
myjob start/running, process 12345
# status myjob
myjob start/running, process 12345
# ps --no-headers -p 12345
# kill 12345
-su: kill: (12345) - No such process

As the ps(1) call shows, the (cat) process is no longer running, but Upstart thinks it is.

Unfortunately, since Upstart will wait forever until it is able to stop the pid (which no longer exits). A manual attempt to either "stop myjob" or "start myjob" will also hang.

The only solution to clear this "stuck job" is to reboot. See [22] and Recovery on Misspecification of expect. Note that this "zombie job" isn't actually causing any problems for Upstart, but it is annoying and potentially confusing seeing it listed in initctl output. It will of course also be consuming a very small amount of memory.

Note however, that if you are working on a development system (hopefully you are whilst developing your job configuration file!), what you can do to keep working is to copy the problematic job configuration file to a new name, ignore the old job entirely and keep working using the new job!

23   Support

The primary sources of support are:

24   References

24.1   Manual Pages

man 5 init
Configuration syntax reference.
man 8 init
Options for running the Upstart init daemon.
man 8 initctl
Explanation of the Upstart control command.
man 7 upstart-events
Comprehensive summary of all "well-known" Upstart system events on Ubuntu.

24.2   Web Sites
Main Ubuntu page for Upstart.
The main Upstart Bazaar project page.
Scotts Original Upstart blog with useful overviews of features and Concepts.
Original Specification.

25   Answers to Test

[2]initctl show-config -e. See initctl show-config.
[3]Job would start "as early as possible": when the startup event is emitted (see Startup Process). It would also be run if the confusingly-named job called "stopped" begun to start (see Starting a Job). It would also be run again if the also confusingly-named job "started" begun to stop (see Stopping a Job). The example chose names that were designed to be confusing. Clearly, in reality you should only create jobs with sensible names that refer to the application they run.
[4]Three times.
[5]Yes. However, it appears that the person who specified this condition failed to read init(5) since it probably won't do what they expect!
[6]The not event.
[7]Here, foo is the value of the first positional environment variable specified by the not event. Upstart treats it as a value since no equals sign is present. This is a convenience since it allows for a more compact (and at times) natural way to specify the start on condition. For example, rather than having to specify "start on started JOB=foo" you can specify the more natural start on started foo. For full details see init(5).
[8]You could trigger the job to start by calling "initctl emit not BLAH_BOING_WOBBLE=foo", but you could equally start the job by calling "initctl emit not FIRST_PARAM=foo".
[9]/tmp is not mounted.

Short answer: "/usr/bin/myapp" will never run. Long answer: This job attempts to only start myapp if it is not disabled by checking its configuration file. However, there are two fatal flaws here:

A corrected version of the Job Configuration File is:

start on runlevel [2345]

env CONFIG=/etc/default/myapp

expect fork

  [ -f "$CONFIG" ] || stop && exit 0
  enabled=$(grep ENABLED=1 $CONFIG || :)
  [ -z "$enabled" ] && exit 0
end script

exec /usr/bin/myapp

Or, if you need to pass options from the config file to the daemon, you could say:

start on runlevel [2345]

env CONFIG=/etc/default/myapp

expect fork

  [ -f "$CONFIG" ] || stop && exit 0
  enabled=$(grep ENABLED=1 $CONFIG || :)
  [ -z "$enabled" ] && exit 0
end script

  exec myapp $MYAPP_OPTIONS
end script

Note how the config file is sourced in the script section and how we specify the shell keyword exec to ensure no sub-shell is created (thus allowing Upstart to track the correct PID).

26   Footnotes

[11]Recall that Upstart has no knowledge of disks whatsoever. In Ubuntu, it relies upon mountall (debian-and-ubuntu-specific) to handle mounting of disks.
[12]Note the method for obtaining the PID of the instance of Upstart running in the LXC container assumes only one other container is running.

Commands to be run as root directly for clarity. However, you should consider using sudo(8) rather than running a root shell. Due to the way sudo works, you have to modify your behaviour slightly. For example, rather than running the following in a root shell:

# echo hello > /tmp/root.txt

You would instead run the command below in a non-root shell:

$ echo hello | sudo tee /tmp/root.txt

Note that you should not use sudo within a job. See Changing User.

[15]If there is a script or exec section and this process is running, state will be pre-stop, else it will be stopping.
[16]Note that the exec line is taken directly from the org.freedesktop.ConsoleKit.service file.
[17]Upstart was written specifically for Ubuntu, although this does not mean that it cannot run on any other Linux-based system. Upstart was first introduced into Ubuntu in release 6.10 ("Edgy Eft"). See
[18]This section of the document contains Ubuntu-specific examples of events. Other operating systems which use Upstart may not implement the same behaviour.
[19](1, 2) This job is not actually available in Ubuntu yet, but is expected to be added early in the 11.10 development cycle.
[20](1, 2) Note that pre-stop does not behave in the same manner as other script sections. See bug 703800 (
[21]For status on chroot support, see bugs 430224 and 728531: - -
[25]A series of blog posts by Scott James Remnant gives further details on events and how they are used. See [26], [27], and [28].
[26](1, 2)
[27](1, 2)
[29](1, 2)
[32]Ubuntu will kill any jobs still running at system shutdown using /etc/init.d/sendsigs.
[33]Note that there is no "startup" job (and hence no /etc/init/startup.conf file).
[34]It is worth noting that Unix and Linux systems are confined by standards to the runlevels specified in the Runlevels section. However, in principle Upstart allows any number of runlevels.

27   Colophon

Copyright:Copyright © 2011-2014, Canonical Ltd. All Rights Reserved. This work is licensed under the Creative Commons Attribution-Share Alike 3.0 Unported License. To view a copy of this license, visit or send a letter to Creative Commons, 171 Second Street, Suite 300, San Francisco, California, 94105, USA.V
Organization:Canonical Ltd.

28   Appendices

28.1   Ubuntu Well-Known Events ()

The information in this section is taken from the upstart-events(7) manual page.


   upstart-events — Well-known Upstart events summary

Event Summary

   This manual page summarises well-known events generated by Upstart running both as the init(8)
   daemon (process ID 1) and a Session Init (process that supervises a user session).

   It is not an exhaustive list of all possible events, but rather details a standard set of
   events expected to be generated on any Ubuntu system running Upstart.

   The primary tables, Table 1 and Table 2, encode the well-known system and session events
   respectively, along with the type of each event (listed in Table 3), the emitter of the event
   (see Table 4) and the approximate time at which the event could be generated. Additionally,
   the Note column indexes into Table 5 for further details on a particular event.

   See init(8) for a table summarising job goals and possible state transitions.

   Note that some events listed in Table 1 may be available to session jobs (depending on when
   the Session Init starts). Those events that are available will be prefixed with :sys:. See
   upstart-event-bridge(8) for further details.

   The Ref (Reference) column is used to refer to individual events succinctly in the Time

   Note that the ''<'' and ''>'' characters in the Time column denote that the event in the Event
   column occurs respectively before or after the event specified in the Time column (for
   example, the mounting(7) event occurs "at some time" after the startup(7) event, and the
   virtual-filesystems(7) event occurs after the last mounted(7) event relating to a virtual
   filesystem has been emitted).

   For further details on events, consult the manual pages and the system job configuration
   files, usually located in /etc/init.

   Table 1. Table 1: Well-Known System Events Summary.

   | Ref |            Event             | Type | Emit |            Time            | Note |
   |     | all-swaps                    |  S   |  M   | > (5)                      |      |
   |     | control-alt-delete(7)        |  S   |  A   | > (5)                      |  A   |
   |     | container                    |  S   |  C   | > /run mounted             |  Q   |
   |     | dbus-activation              |  S   |  B   | > D-Bus client request     |      |
   |     | deconfiguring-networking     |  H   |  V   | < non-local IFs down       |  P   |
   |     | desktop-session-start        |  H   |  D   | > X(7) session created     |  B   |
   |     | desktop-shutdown             |  H   |  D   | > X(7) session ended       |  O   |
   |     | device-not-ready             |  H   |  M   | > (2)                      |  N   |
   |     | drm-device-added             |  S   |  U   | > (5)                      |  C   |
   |     | failsafe-boot                |  S   |  X   | > (7) and local IF         |  S   |
   |     | file                         |  S   |  K   | > (1)                      |  U   |
   |  7  | filesystem                   |  S   |  M   | After last (1)             |  D   |
   |     | graphics-device-added        |  S   |  U   | > (5)                      |  C   |
   |     | keyboard-request(7)          |  S   |  A   | > (5)                      |  E   |
   |     | local-filesystems(7)         |  S   |  M   | > (6)                      |      |
   |     | login-session-start          |  H   |  D   | < DM running               |  F   |
   |  1  | mounted(7)                   |  H   |  M   | > associated (2)           |  G   |
   |  2  | mounting(7)                  |  H   |  M   | > (5)                      |  H   |
   |  3  | net-device-added             |  S   |  U   | > (5)                      |  C   |
   |     | net-device-changed           |  S   |  U   | > (5)                      |  C   |
   |     | net-device-down              |  S   |  F   | < (4)                      |  C   |
   |  4  | net-device-removed           |  S   |  U   | > (5)                      |  C   |
   |     | net-device-up                |  S   | F,N  | > (3)                      |  C   |
   |     | not-container                |  S   |  C   | > /run mounted             |  Q   |
   |     | power-status-changed(7)      |  S   |  I   | > (5)                      |  I   |
   |     | recovery                     |  S   |  G   | Boot (<5)                  |  R   |
   |     | remote-filesystems(7)        |  S   |  M   | > (6)                      |      |
   |     | rotate-logs                  |  H   |  R   | daily                      |  Y   |
   |     | runlevel(7)                  |  M   |  T   | > (7) + (8)                |      |
   |     | socket(7)                    |  S   |  S   | > socket connection        |  X   |
   |  5  | startup(7)                   |  S   |  I   | Boot                       |  J   |
   |     | started(7)                   |  S   |  I   | > job started              |  K   |
   |     | starting(7)                  |  H   |  I   | < job starts               |  K   |
   |  8  | static-network-up            |  S   |  N   | > last static IF up        |      |
   |     | stopped(7)                   |  S   |  I   | > job stopped              |  K   |
   |     | stopping(7)                  |  H   |  I   | < job stops                |  K   |
   |     | unmounted-remote-filesystems |  H   |  V   | > last remote FS unmounted |  L   |
   |  6  | virtual-filesystems(7)       |  S   |  M   | > last virtual FS (1)      |  M   |

   Key: ''DM'' is an abbreviation for Display Manager. ''FS'' is an abbreviation for filesystem.
   ''IF'' is an abbreviation for Network Interface.

   Table 2. Table 2: Well-Known User Events Summary.

   | Ref |      Event       | Type | Emit |              Time               | Note |
   |     | dbus             |  S   |  L   | > (1)                           |  W   |
   |     | dconf            |  S   |  O   | > (1)                           |      |
   |     | desktop-end(7)   |  S   |  J   | < (2)                           |      |
   |     | desktop-start(7) |  H   |  J   | > (3)                           |      |
   |     | file             |  S   |  K   | > (1)                           |  U   |
   |  2  | session-end(7)   |  M   |  I   | < Session Init end              |      |
   |  1  | startup(7)       |  S   |  I   | > Session Init start            |  J   |
   |     | :sys:*           |  S   |  E   | > upstart-event-bridge(8) start |      |
   |     | :sys:restarted   |  S   |  E   | > upstart-event-bridge(8) start |  V   |
   |  3  | xsession         |  M   |  H   | > (1)                           |  T   |

   Table 3. Table 3: Event Types.

   | Ref | Event Type | Notes                                                           |
   |  H  | Hook       | Blocking. Waits for events that start on or stop on this event. |
   |  M  | Method     | Blocking task.                                                  |
   |  S  | Signal     | Non-blocking.                                                   |

   Table 4. Table 4: Event Emitters.

   | Ref | Emitter                          | Notes                           |
   |  A  | System Administrator (initiator) | Technically emitted by init(8). |
   |  B  | dbus-daemon(1)                   | Run with "--activation=upstart" |
   |  C  | container-detect job             |                                 |
   |  D  | Display Manager                  | e.g. lightdm/gdm/kdm/xdm.       |
   |  E  | upstart-event-bridge(8)          |                                 |
   |  F  | ifup(8) or ifdown(8)             | See /etc/network/.              |
   |  G  | bootloader or initramfs          |                                 |
   |  H  | xsession-init session job        |                                 |
   |  I  | init(8)                          | Either PID 1 or a Session Init. |
   |  J  | job that starts desktop          | gnome-session job for Ubuntu.   |
   |  K  | upstart-file-bridge(8)           | See file-event(7).              |
   |  L  | upstart-dbus-bridge(8)           | See dbus-event(7).              |
   |  M  | mountall(8)                      |                                 |
   |  N  | network-interface job            |                                 |
   |  O  | upstart-dconf-bridge(8)          | See dconf-event(7).             |
   |  R  | /etc/cron.daily/upstart          | See cron(8).                    |
   |  S  | upstart-socket-bridge(8)         | See socket-event(7).            |
   |  T  | telinit(8), shutdown(8)          |                                 |
   |  U  | upstart-udev-bridge(8)           |                                 |
   |  V  | System V init system             |                                 |
   |  X  | failsafe job                     |                                 |

   Table 5. Table 5: Event Summary Notes.

   | Note | Detail                                                                              |
   |  A   | Requires administrator to press Control-Alt-Delete key combination on the console.  |
   |  B   | Event generated when user performs graphical login.                                 |
   |      | These are specific examples. upstart-udev-bridge(8) will emit events which match    |
   |  C   | the pattern, "S-device-A" where ''S'' is the udev subsystem and ''A'' is the udev   |
   |      | action. See udev(7) and for further details. If you have sysfs mounted, you can     |
   |      | look in /sys/class/ for possible values for subsystem.                              |
   |  D   | Note this is in the singular - there is no ''filesystems'' event.                   |
   |  E   | Emitted when administrator presses Alt-UpArrow key combination on the console.      |
   |  F   | Denotes Display Manager running (about to be displayed), but no users logged in     |
   |      | yet.                                                                                |
   |  G   | Generated for each mount that completes successfully.                               |
   |  H   | Emitted when mount attempt for single entry from fstab(5) for any filesystem type   |
   |      | is about to begin.                                                                  |
   |  I   | Emitted when Upstart receives the SIGPWR signal.                                    |
   |  J   | Initial event (system or Session Init).                                             |
   |  K   | Although the events are emitted by init(8), the instigator may be initctl(8) if a   |
   |      | System Administrator has manually started or stopped a job.                         |
   |  L   | /etc/init/                                                             |
   |  M   | Emitted when all virtual filesystems (such as /proc) mounted.                       |
   |  N   | Emitted when the --dev-wait-time timeout is exceeded for mountall(8). This defaults |
   |      | to 30 seconds.                                                                      |
   |  O   | Emitted when the X(7) display manager exits at shutdown or reboot, to hand off to   |
   |      | the shutdown splash manager.                                                        |
   |  P   | Emitted by /etc/init.d/networking just prior to stopping all non-local network      |
   |      | interfaces.                                                                         |
   |  Q   | Either ''container'' or ''not-container'' is emitted (depending on the              |
   |      | environment), but not both.                                                         |
   |      | Emitted by either the initramfs or bootloader (for example grub) as the initial     |
   |  R   | event (rather than startup(7)) to denote the system has booted into recovery mode.  |
   |      | If recovery was successful, the standard startup(7) event is then emitted, allowing |
   |      | the system to boot as normal.                                                       |
   |      | Emitted to indicate the system has failed to boot within the expected time. This    |
   |  S   | event will trigger other jobs to forcibly attempt to bring the system into a usable |
   |      | state.                                                                              |
   |  T   | Only emitted for a graphical session.                                               |
   |  U   | See file-event(7).                                                                  |
   |  V   | This is a pseudo-system event emitted directly by the upstart-event-bridge(8).      |
   |  W   | Emitted by Session Init only since bridge not run as system job by default.         |
   |  X   | See socket-event(7).                                                                |
   |  Y   | Used primary by Session Job /usr/share/upstart/sessions/logrotate.conf.             |


   Manual page written by James Hunt <<[1]>>


   Report bugs at <>


   Copyright © 2011-2013 Canonical Ltd.

   This is free software; see the source for copying conditions. There is NO warranty; not even


   all-swaps(7), control-alt-delete(7), dbus-daemon(1), dbus-event(7), dconf-event(7),
   file-event(7), filesystem(7), ifdown(8) ifup(8) init(5), init(8), initctl(8),
   keyboard-request(7), local-filesystems(7), mountall(8), mounted(7), mounting(7),
   power-status-changed(7), remote-filesystems(7), runlevel(7), shutdown(8), socket-event(7),
   started(7), starting(7), startup(7), stopped(7), stopping(7), telinit(8),
   upstart-dbus-bridge(8), upstart-dconf-bridge(8), upstart-event-bridge(8),
   upstart-file-bridge(8), upstart-local-bridge(8), upstart-socket-bridge(8),
   upstart-udev-bridge(8), virtual-filesystems(7).


   Visible links