README.md 17 KB
Dinit
v0.9.0 (development release)
This is the README for Dinit, the service manager and init system. It is intended to provide an overview; For full documentation please check the manual pages. The impatient may wish to check out the getting started guide.
Contents
Introduction
Dinit is a service supervisor with dependency support which can also act as the system "init" program. It was created with the intention of providing a portable init system with dependency management, that was functionally superior to many extant inits. On Linux it can serve as a lighter-weight alternative to Systemd. Development goals include clean design, robustness, portability, usability, and avoiding feature bloat (whilst still handling a variety of use cases).
To elaborate, Dinit can launch multiple services in parallel, with dependency management (i.e. if one service's operation depends on another, the latter service will be started first). It can monitor the process corresponding to a service, and re-start it if it dies, and it can do this in an intelligent way, first "rolling back" all dependent services, and restarting them when their dependencies are satisfied. However, the precise nature of dependency relations between services is highly configurable. The dinitctl tool can be used to start or stop services and check their state (by issuing commands to the "dinit" daemon).
Dinit is designed to run as either as a system service manager (runs as root, uses system paths for configuration etc) or a user process (runs as a user, uses paths in the user's home directory by default).
Dinit is designed to work on POSIXy operating systems such as Linux and OpenBSD. It is written in C++ and uses the Dasynq event handling library, which was written especially to support Dinit. (Note that a copy of Dasynq is bundled with Dinit, so a separate copy is not required for compilation; however, the bundled copy does not include the documentation or test suite).
See doc/COMPARISON for a comparison of Dinit with similar software packages.
Dinit is licensed under the Apache License, version 2.0. A copy of this license can be found in the LICENSE file.
Dinit was written by Davin McCall davmac@davmac.org.
See BUILD.txt for information on how to build Dinit.
Configuring services
Service types
A "service" is nominally a persistent process or system state. The two main types of service are a process service (represented by a an actual process) and a scripted service (which is started and stopped by running a process - often a shell script - to completion). There are also bgprocess services and internal services.
Many programs that you might want to run under Dinit's supervision can run either "in the foreground" or as a daemon ("in the background"), and the choice is dictated by a command line switch (for instance the -D and -F switches to Samba's "smbd"). Although it might seem counterintuitive, the "foreground" mode should be used for programs registered as process services in Dinit; this allows Dinit to monitor the process.
Process services are attractive due to the ease of monitoring (and restarting) the service. After starting a process, there will often be a short delay before the process sets itself up, starts listening on sockets, etc; during this time any other process (including one from a service configured as a dependent) which tries to contact it will not be able to do so. In practice, this is not usually an issue (and external solutions, like D-Bus, do exist), but Dinit does support startup notification (compatible with S6) to circumvent the problem. With startup notification configured - assuming it is supported by the process - dependent services will not be started until the service is running properly. If startup notification is not configured, Dinit assumes a process service is successfully started as soon as the process is launched.
A scripted service has separate commands for startup and (optional) shutdown. Scripted services can be used for tasks such as mounting file systems that don't need a persistent process, and in some cases can be used for daemon processes (although Dinit will not be able to supervise a process that is registered as a scripted service).
A bgprocess service is a mix between a process service and a scripted service. A command is used to start the service, and once started, the process ID is expected to be available in a file which Dinit can then read. Many existing daemons can operate in this way. The process can only be supervised if Dinit runs as the system "init" (PID 1), or can otherwise mark itself as a subreaper (which is possible on Linux, FreeBSD and DragonFlyBSD) - otherwise Dinit can not reliably know when the process has terminated.
(Note, use of bgprocess services type requires care. The file from which the PID is read is trusted; Dinit may send signals to the specified PID. It should not be possible for unauthorised users to modify the file contents!)
An internal service is just a placeholder service that can be used to describe a set of dependencies. An internal service has no corresponding process.
Service description files
Dinit discovers services by reading service description files. These files
reside in a directory (/etc/dinit.d is the default "system" location, with
/usr/local/lib/dinit.d and /lib/dinit.d also searched; the default user
location is $HOME/dinit.d) and the name of a service description file
matches the name of the service they configure.
For example, a service named "mysql" might be configured via the service description
file named /etc/dinit.d/mysql. Service descriptions are loaded lazily, as needed
by Dinit; so, this service description file will usually be read when the mysql
service is first staretd.
(An example of a complete set of system service descriptions can be found in the doc/linux/services directory).
A service description file has a textual format and consists of a number of
parameter settings. Settings in the file are denoted as a parameter name followed
by either an equal sign or colon and then the parameter value (all on the same line).
Comments begin with a hash mark (#) and extend to the end of the line (they
must be separated from setting values by at least one whitespace character).
Parameter values are interpreted literally, except that:
- whitespace is collapsed to a single space
- double quotes can be used around all or part(s) of a parameter to prevent whitespace collapse and interpretation of special characters
- backslash can be used to 'escape' the next character, preventing any special meaning from being associated with it. It can be used to include non-collapsing whitespace, double-quote marks, and backslashes in the parameter value.
Some examples of the available parameters are:
type = process | bgprocess | scripted | internal
command = ...
stop-command = ...
run-as = (user-id)
restart = (boolean)
smooth-recovery = (boolean)
logfile = ...
pid-file = ...
options = ...
depends-on = (service name)
depends-ms = (service name)
waits-for = (service name)
Descriptions of individual parameters follows:
command = (external script or executable, and arguments)
For a 'process' service, this is the process to run. For a 'scripted' service, this command is run to start the service.
stop-command = (external script or executable, and arguments)
For a 'scripted' service, this command is run to stop the service.
run-as = (user-id)
Specifies which user to run the process(es) for this service as. The group id for the process will also be set to the primary group of the specified user.
restart = yes | true | no | false
Specifies whether the service should automatically restart if it becomes stopped (for any reason, including being explicitly requested to stop). Only active services will restart automatically.
smooth-recovery = yes | true | no | false
For process services only. Specifies that, should the process die, it can be restarted without bringing the service itself down. This means that any dependent services do not need to be stopped/restarted. Such recovery happens regardless of the "restart" setting (if smooth-recovery is enabled, the service does not reach the stopped state when the process terminates unexpectedly).
logfile = (log file path)
Specifies the log file for the service. Output from the service process will go this file.
pid-file = (path to file)
For "bgprocess" type services only; specifies the path of the file where daemon will write its process ID before detaching.
depends-on = (service name)
This service depends on the named service. Starting this service will start the named service; the command to start this service will not be executed until the named service has started. If the named service is stopped then this service will also be stopped.
depends-ms = (service name)
Indicates a "milestone dependency" on the named service. This service requires the named service to start before it starts itself. Once the named service has started, it remains active due to the dependency, but if it stops for any reason then the dependency link is broken until the next time this service is started.
waits-for = (service name)
When this service is started, wait for the named service to finish starting (or to fail starting) before commencing the start procedure for this service. Starting this service will automatically start the named service.
options = ( no-sigterm | runs-on-console | starts-on-console | start-interruptible ) ...
Specifies various options for this service. Some of the possible options include:
no-sigterm : specifies that the TERM signal should not be send to the
process to terminate it. (Another signal can be specified using
the `termsignal` setting; if no other signal is specified, *no*
signal will be sent).
runs-on-console : specifies that this service uses the console; its input
and output should be directed to the console. A service running
on the console prevents other services from running on the
console (they will queue for the console).
The "interrupt" key (normally control-C) will be active for
process / scripted services that run on the console. Handling
of an interrupt is determined by the service process, but
typically will cause it to terminate.
starts-on-console : specifies that this service uses the console during
service startup. This is implied by runs-on-console, but can
be specified separately for services that need the console
while they start but not afterwards.
This setting is not applicable to regular _process_ services,
but can be used for _scripted_ and _bgprocess_ services. It
allows for interrupting startup via the "interrupt" key
(normally control-C). This is useful to allow filesystem checks
to be interrupted/skipped.
start-interruptible : this service can have its startup interrupted
(cancelled) if it becomes inactive while still starting.
The SIGINT signal will be sent to the process to cancel its
startup. This is meaningful only for _scripted_ and _bgprocess_
services.
Please see the manual page for a full list of service parameters and options.
Running Dinit
Dinit can run as the system "init" - the first process started by the kernel
on boot - which is normally done by linking or copying it to /sbin/init.
This is currently supported only on Linux. It requires having suitable service
descriptions in place and should be attempted only by those comfortable
with low-level system administration and recovery. See doc/linux directory for
more information.
Dinit can also run as a normal process, and can be started in this case by a regular user.
By default, regardless of whether it runs as a system or user process, Dinit
will look for and start the service named "boot". This service should be
configured with dependencies which will cause any other desired services to
start. You can specify alternative services to start via the dinit command
line (consult the man page for more information).
Controlling services
Service hierarchy and states
Services can depend on other services for operation, and so form a dependency hierarchy. Starting a service which depends on another causes that other service to start (and the first service waits until the latter has started before its process is launched and it is itself considered started).
Services are considered active when they are not stopped. Services can also be explicitly marked as active (this normally happens when you explicitly start a service). Finally, a service with an active dependent is also considered active.
If a service stops and becomes inactive (i.e. it is not explicitly marked active and has no active dependents) then any services it depends on will also be marked inactive and stopped unless they have other active dependents, or were explicitly started and marked active.
What this means is that, in general, starting an (inactive, stopped) service and then stopping it will return the system to its prior state - no dependencies which were started automatically will be left running.
Using dinitctl
You can use the "dinitctl" utility to start and stop services. Typical invocations are:
dinitctl start <service-name>
dinitctl stop <service-name>
dinitctl release <service-name>
Note that a start marks the service active, as well as starting it if it is not already started; the opposite of this is actually release, which clears the active mark and stops it if it has no active dependent services.
The stop command by default acts as a release that also forces the service to
stop (although it may then immediately restart, depending on how it and its
dependents are configured). If stopping a service would also require a dependent
service to stop, a warning will be issued; the --force option will be required
to bypass the warning.
When run as root, dinitctl (by default) communicates with the system instance of
Dinit. Otherwise, it communicates with a user (personal) instance. This can be
overridden (using -u or -s for the user or system instance, respectively), but
note that regular users will generally lack the required permission to communicate
with the system instance.
Here is an example command for starting a service:
dinitctl start mysql # start mysql service
For complete details on the command line, use:
dinitctl --help
You can "pin" a service in either the stopped or started state, which prevents it from changing state either due to a dependency/dependent or a direct command:
dinitctl start --pin mysql # start mysql service, pin it as "started"
dinitctl stop mysql # issues stop, but doesn't take effect due to pin
dinitctl unpin mysql # release pin; service will now stop
You can pin a service in the stopped state in order to make sure it doesn't get started accidentally (either via a dependency or directly). You can also use it to temporarily keep stopped a service that would otherwise restart immediately when you stopped it (because it, or a dependent, is configured to restart automatically).
Finally, you can list the state of all loaded services:
dinitctl list
This may result in something like the following:
[{+} ] boot
[{+} ] tty1 (pid: 300)
[{+} ] tty2 (pid: 301)
[{+} ] tty3 (pid: 302)
[{+} ] tty4 (pid: 303)
[{+} ] loginready (has console)
[{+} ] rcboot
[{+} ] filesystems
[{+} ] udevd (pid: 4)
[ {-}] mysql
The above represents a number of started services and one stopped service (mysql). Services transitioning state (starting or stopping) are displayed with an arrow indicating the transition direction:
[{ }<< ] mysql # starting
[ >>{ }] mysql # stopping
The curly brackets indicate the desired state, which may not be the state to which the service is currently transitioning. For example:
[ <<{ }] mysql # starting, but will stop after starting
[{ }>> ] mysql # stopping, but will restart once stopped
Remember that a starting service may be waiting for its dependencies to start, and a stopping service may be waiting for its dependencies to stop.