zdaemon 5.1

Daemon process control library and tools for Unix-based systems

zdaemon is a Unix (Unix, Linux, Mac OS X) Python program that wraps commands to make them behave as proper daemons.

Contents

• Using zdaemon

  • Environment Variables

  • Transcript log

  • Start test program and timeout

  • Reference Documentation

• Change log

  • 5.1 (2024-05-03)

  • 5.0 (2023-05-24)

  • 4.4 (2022-12-02)

  • 4.3 (2018-10-30)

  • 4.2.0 (2016-12-07)

  • 4.1.0 (2015-04-16)

  • 4.0.1 (2014-12-26)

  • 4.0.0 (2013-05-10)

  • 4.0.0a1 (2013-02-15)

  • 3.0.5 (2012-11-27)

  • 3.0.4 (2012-07-30)

  • 3.0.3 (2012-07-10)

  • 3.0.2 (2012-07-10)

  • 3.0.1 (2012-06-08)

  • 3.0.0 (2012-06-08)

  • 2.0.6 (2012-06-07)

  • 2.0.5 (2012-06-07)

  • 2.0.4 (2009-04-20)

  • 2.0.3 (2009-04-11)

    • Bugs Fixed

  • 2.0.2 (2008-04-05)

  • 2.0.1 (2007-10-31)

  • 2.0.0 (2007-07-19)

  • 2.0a6 (2007-01-11)

  • 2.0a3, 2.0a4, 2.0a5 (2007-01-10)

  • 2.0a2 (2007-01-10)

  • 2.0a1 (2006-12-21)

    • Bugs Fixed

    • New Features

  • 1.4a1 (2005-11-21)

  • 1.4a1 (2005-11-05)

  • After some unknown release(???)

  • After 1.1

  • 1.1 (2005-06-09)

  • To-Dos

Using zdaemon

zdaemon provides a script, zdaemon, that can be used to run other programs as POSIX (Unix) daemons. (Of course, it is only usable on POSIX-complient systems.)

Using zdaemon requires specifying a number of options, which can be given in a configuration file, or as command-line options. It also accepts commands teling it what do do. The commands are:

start

Start a process as a daemon

stop

Stop a running daemon process

restart

Stop and then restart a program

status

Find out if the program is running

foreground or fg

Run a program

kill signal

Send a signal to the daemon process

reopen_transcript

Reopen the transcript log. See the discussion of the transcript log below.

help command

Get help on a command

Commands can be given on a command line, or can be given using an interactive interpreter.

Let’s start with a simple example. We’ll use command-line options to run the echo command:

sh> ./zdaemon -p ‘echo hello world’ fg echo hello world hello world

Here we used the -p option to specify a program to run. We can specify a program name and command-line options in the program command. Note, however, that the command-line parsing is pretty primitive. Quotes and spaces aren’t handled correctly. Let’s look at a slightly more complex example. We’ll run the sleep command as a daemon :)

sh> ./zdaemon -p ‘sleep 100’ start . . daemon process started, pid=819

This ran the sleep daemon. We can check whether it ran with the status command:

sh> ./zdaemon -p ‘sleep 100’ status program running; pid=819

We can stop it with the stop command:

sh> ./zdaemon -p ‘sleep 100’ stop . . daemon process stopped

sh> ./zdaemon -p ‘sleep 100’ status daemon manager not running Failed: 3

Normally, we control zdaemon using a configuration file. Let’s create a typical configuration file:

<runner>
  program sleep 100
</runner>

Now, we can run with the -C option to read the configuration file:

sh> ./zdaemon -Cconf start . . daemon process started, pid=1136

If we list the directory:

sh> ls conf zdaemon zdsock

We’ll see that a file, zdsock, was created. This is a unix-domain socket used internally by ZDaemon. We’ll normally want to control where this goes.

sh> ./zdaemon -Cconf stop . . daemon process stopped

Here’s an updated configuration:

<runner>
  program sleep 100
  socket-name /tmp/demo.zdsock
</runner>

Now, when we run zdaemon:

sh> ./zdaemon -Cconf start . . daemon process started, pid=1139

sh> ls conf zdaemon

The socket file is created in the given directory.

sh> ./zdaemon -Cconf stop . . daemon process stopped

In the example, we included a command-line argument in the program option. We can also provide options on the command line:

<runner>
  program sleep
  socket-name /tmp/demo.zdsock
</runner>

Then we can pass the program argument on the command line:

sh> ./zdaemon -Cconf start 100 . . daemon process started, pid=1149

sh> ./zdaemon -Cconf status program running; pid=1149

sh> ./zdaemon -Cconf stop . . daemon process stopped

Environment Variables

Sometimes, it is necessary to set environment variables before running a program. Perhaps the most common case for this is setting LD_LIBRARY_PATH so that dynamically loaded libraries can be found.

<runner>
  program env
  socket-name /tmp/demo.zdsock
</runner>
<environment>
  LIBRARY_PATH /home/foo/lib
  HOME /home/foo
</environment>

Now, when we run the command, we’ll see out environment settings reflected:

sh> ./zdaemon -Cconf fg env USER=jim HOME=/home/foo LOGNAME=jim USERNAME=jim TERM=dumb PATH=/home/jim/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin EMACS=t LANG=en_US.UTF-8 SHELL=/bin/bash EDITOR=emacs LIBRARY_PATH=/home/foo/lib

Transcript log

When zdaemon run a program in daemon mode, it disconnects the program’s standard input, standard output, and standard error from the controlling terminal. It can optionally redirect the output to standard error and standard output to a file. This is done with the transcript option. This is, of course, useful for logging output from long-running applications.

Let’s look at an example. We’ll have a long-running process that simple tails a data file:

>>> f = open('data', 'w', 1)
>>> import os
>>> _ = f.write('rec 1\n'); f.flush(); os.fsync(f.fileno())

Now, here’s out zdaemon configuration:

<runner>
  program tail -f data
  transcript log
</runner>

Now we’ll start:

sh> ./zdaemon -Cconf start . . daemon process started, pid=7963

After waiting a bit, if we look at the log file, it contains the tail output:

>>> with open('log') as file:
...     file.read()
'rec 1\n'

We can rotate the transcript log by renaming it and telling zdaemon to reopen it:

>>> import os
>>> os.rename('log', 'log.1')

If we generate more output:

>>> _ = f.write('rec 2\n'); f.flush(); os.fsync(f.fileno())

The output will appear in the old file, because zdaemon still has it open:

>>> with open('log.1') as file:
...     file.read()
'rec 1\nrec 2\n'

Now, if we tell zdaemon to reopen the file:

sh> ./zdaemon -Cconf reopen_transcript

and generate some output:

>>> _ = f.write('rec 3\n'); f.flush(); os.fsync(f.fileno())

the output will show up in the new file, not the old:

>>> with open('log') as file:
...     file.read()
'rec 3\n'

>>> with open('log.1') as file:
...     file.read()
'rec 1\nrec 2\n'

Close files and clean up:

>>> f.close()

sh> ./zdaemon -Cconf stop . . daemon process stopped

Start test program and timeout

Normally, zdaemon considers a process to have started when the process itself has been created. A process may take a while before it is truly up and running. For example, a database server or a web server may take time before they’re ready to accept requests.

You can optionally supply a test program, via the start-test-program configuration option, that is called repeatedly until it returns a 0 exit status or until a time limit, start-timeout, has been reached.

Reference Documentation

The following options are available for use in the runner section of configuration files and as command-line options.

program

Command-line option: -p or –program

This option gives the command used to start the subprocess managed by zdaemon. This is currently a simple list of whitespace-delimited words. The first word is the program file, subsequent words are its command line arguments. If the program file contains no slashes, it is searched using $PATH. (Note that there is no way to to include whitespace in the program file or an argument, and under certain circumstances other shell metacharacters are also a problem.)

socket-name

Command-line option: -s or –socket-name.

The pathname of the Unix domain socket used for communication between the zdaemon command-line tool and a daemon-management process. The default is relative to the current directory in which zdaemon is started. You want to specify an absolute pathname here.

This defaults to “zdsock”, which is created in the directory in which zdrun is started.

daemon

Command-line option: -d or –daemon.

If this option is true, zdaemon runs in the background as a true daemon. It forks a child process which becomes the subprocess manager, while the parent exits (making the shell that started it believe it is done). The child process also does the following:

• if the directory option is set, change into that directory

• redirect stdin, stdout and stderr to /dev/null

• call setsid() so it becomes a session leader

• call umask() with specified value

The default for this option is on by default. The command-line option therefore has no effect. To disable daemon mode, you must use a configuration file:

<runner>
  program sleep 1
  daemon off
</runner>

directory

Command-line option: -z or –directory.

If the daemon option is true (default), this option can specify a directory into which zdrun.py changes as part of the “daemonizing”. If the daemon option is false, this option is ignored.

backoff-limit

Command-line option: -b or –backoff-limit.

When the subprocess crashes, zdaemon inserts a one-second delay before it restarts it. When the subprocess crashes again right away, the delay is incremented by one second, and so on. What happens when the delay has reached the value of backoff-limit (in seconds), depends on the value of the forever option. If forever is false, zdaemon gives up at this point, and exits. An always-crashing subprocess will have been restarted exactly backoff-limit times in this case. If forever is true, zdaemon continues to attempt to restart the process, keeping the delay at backoff-limit seconds.

If the subprocess stays up for more than backoff-limit seconds, the delay is reset to 1 second.

This defaults to 10.

forever

Command-line option: -f or –forever.

If this option is true, zdaemon will keep restarting a crashing subprocess forever. If it is false, it will give up after backoff-limit crashes in a row. See the description of backoff-limit for details.

This is disabled by default.

exit-codes

Command-line option: -x or –exit-codes.

This defaults to 0,2.

If the subprocess exits with an exit status that is equal to one of the integers in this list, zdaemon will not restart it. The default list requires some explanation. Exit status 0 is considered a willful successful exit; the ZEO and Zope server processes use this exit status when they want to stop without being restarted. (Including in response to a SIGTERM.) Exit status 2 is typically issued for command line syntax errors; in this case, restarting the program will not help!

NOTE: this mechanism overrides the backoff-limit and forever options; i.e. even if forever is true, a subprocess exit status code in this list makes zdaemon give up. To disable this, change the value to an empty list.

start-test-program

A command that tests whether the program is up and running. The command should exit with a zero exit statis if the program is running and with a non-zero status otherwise.

start-timeout

Command-line option: -T or –start-timeout.

If the program takes more than start-timeout seconds to start, then an error is printed and the control script will exit with a non-zero exit status.

stop-timeout

This defaults to 300 seconds (5 minutes).

When a stop command is issued, a SIGTERM signal is sent to the process. zdaemon waits for stop-timeout seconds for the process to gracefully exit. If the process doesn’t exit in that time, a SIGKILL signal is sent.

user

Command-line option: -u or –user.

When zdaemon is started by root, this option specifies the user as who the the zdaemon process (and hence the daemon subprocess) will run. This can be a user name or a numeric user id. Both the user and the group are set from the corresponding password entry, using setuid() and setgid(). This is done before zdaemon does anything else besides parsing its command line arguments.

NOTE: when zdaemon is not started by root, specifying this option is an error. (XXX This may be a mistake.)

XXX The zdaemon event log file may be opened before setuid() is called. Is this good or bad?

umask

Command-line option: -m or –umask.

When daemon mode is used, this option specifies the octal umask of the subprocess.

default-to-interactive

If this option is true, zdaemon enters interactive mode when it is invoked without a positional command argument. If it is false, you must use the -i or –interactive command line option to zdaemon to enter interactive mode.

This is enabled by default.

logfile

Command-line option: -l or –logfile.

This option specifies a log file that is the default target of the “logtail” zdaemon command.

NOTE: This is NOT the log file to which zdaemon writes its logging messages! That log file is specified by the <eventlog> section described below.

transcript

Command-line option: -t or –transcript.

The name of a file in which a transcript of all output from the command being run will be written to when daemonized.

If not specified, output from the command will be discarded.

This only takes effect when the “daemon” option is enabled.

prompt

The prompt shown by the controller program. The default must be provided by the application.

(Note that a few other options are available to support old configuration files, but aren’t needed any more and can generally be ignored.)

In addition to the runner section, you can use an eventlog section that specified one or more logfile subsections:

<eventlog>
  <logfile>
    path /var/log/foo/foo.log
  </logfile>

  <logfile>
    path STDOUT
  </logfile>
</eventlog>

In this example, log output is sent to a file and to standard out. Log output from zdaemon usually isn’t very interesting but can be handy for debugging.

Change log

5.1 (2024-05-03)

• Add support for Python 3.12.

• Fix SIGCHLD/wait raise condition associated with the start-test-program option. For details see #33.

5.0 (2023-05-24)

• Drop support for Python 2.7, 3.5, 3.6.

4.4 (2022-12-02)

• Add support for Python 3.8, 3.9, 3.10, 3.11.

• Drop support for Python 3.4.

• Drop support for python setup.py test to run the tests. (#23)

• Drop support for installing this package without having setuptools.

4.3 (2018-10-30)

• Add support for Python 3.6 and 3.7.

• Drop support for Python 3.3.

4.2.0 (2016-12-07)

• Add support for Python 3.5.

• Drop support for Python 2.6 and 3.2.

4.1.0 (2015-04-16)

• Add --version command line option (fixes https://github.com/zopefoundation/zdaemon/issues/4).

• kill now accepts signal names, not just numbers (https://github.com/zopefoundation/zdaemon/issues/11).

• Restore logreopen as an alias for kill USR2 (removed in version 3.0.0 due to lack of tests): https://github.com/zopefoundation/zdaemon/issues/10.

• Make logreopen also reopen the transcript log: https://github.com/zopefoundation/zdaemon/issues/9.

• Reopen event log on logreopen or reopen_transcript: https://github.com/zopefoundation/zdaemon/issues/8.

• Help message for reopen_transcript (https://github.com/zopefoundation/zdaemon/issues/5).

• Fix race condition where stop would be ignored if the daemon manager was waiting before respawning a crashed program. https://github.com/zopefoundation/zdaemon/issues/13.

• Partially fix delayed deadlock when the transcript file runs into a full disk (https://github.com/zopefoundation/zdaemon/issues/1).

• Fix test suite leaving stale processes behind (https://github.com/zopefoundation/zdaemon/issues/7).

4.0.1 (2014-12-26)

• Add support for PyPy. (PyPy3 is pending release of a fix for: https://bitbucket.org/pypy/pypy/issue/1946)

• Add support for Python 3.4.

• Add -t/--transcript command line option.

• zdaemon can now be invoked as a module as in python -m zdaemon ...

4.0.0 (2013-05-10)

• Add support for Python 3.2.

4.0.0a1 (2013-02-15)

• Add tox support and MANIFEST.in for proper releasing.

• Add Python 3.3 support.

• Drop Python 2.4 and 2.5 support.

3.0.5 (2012-11-27)

• Fixed: the status command didn’t return a non-zero exit status when the program wasn’t running. This made it impossible for other software (e.g. Puppet) to tell if a process was running.

3.0.4 (2012-07-30)

• Fixed: The start command exited with a zero exit status even when the program being started failed to start (or exited imediately).

3.0.3 (2012-07-10)

• Fixed: programs started with zdaemon couldn’t, themselves, invoke zdaemon.

3.0.2 (2012-07-10)

Fail :(

3.0.1 (2012-06-08)

• Fixed:

  The change in 2.0.6 to set a user’s supplemental groups broke common configurations in which the effective user was set via su or sudo -u prior to invoking zdaemon.

  Now, zdaemon doesn’t set groups or the effective user if the effective user is already set to the configured user.

3.0.0 (2012-06-08)

• Added an option, start-test-program to supply a test command to test whether the program managed by zdaemon is up and operational, rather than just running. When starting a program, the start command doesn’t return until the test passes. You could, for example, use this to wait until a web server is actually accepting connections.

• Added a start-timeout option to error if a program takes too long to start. This is especially useful in combination with the start-test-program option.

• Added an option, stop-timeout, to control how long to wait for a graceful shutdown.

  Previously, this was controlled by backoff-limit, which didn’t make much sense.

• Several undocumented, untested, and presumably unused features were removed.

2.0.6 (2012-06-07)

• Fixed: When the user option was used to run as a particular user, supplemental groups weren’t set to the user’s supplemental groups.

2.0.5 (2012-06-07)

(Accidental release. Please ignore.)

2.0.4 (2009-04-20)

• Version 2.0.3 broke support for relative paths to the socket (-s option and socket-name parameter), now relative paths work again as in version 2.0.2.

• Fixed change log format, made table of contents nicer.

• Fixed author’s email address.

• Removed zpkg stuff.

2.0.3 (2009-04-11)

• Added support to bootstrap on Jython.

• If the run directory does not exist it will be created. This allow to use /var/run/mydaemon as run directory when /var/run is a tmpfs (LP #318118).

Bugs Fixed

• No longer uses a hard-coded file name (/tmp/demo.zdsock) in unit tests. This lets you run the tests on Python 2.4 and 2.5 simultaneously without spurious errors.

• make -h work again for both runner and control scripts. Help is now taken from the __doc__ of the options class users by the zdaemon script being run.

2.0.2 (2008-04-05)

Bugs Fixed

• Fixed backwards incompatible change in handling of environment option.

2.0.1 (2007-10-31)

Bugs Fixed

• Fixed test renormalizer that did not work in certain cases where the environment was complex.

2.0.0 (2007-07-19)

• Final release for 2.0.0.

2.0a6 (2007-01-11)

Bugs Fixed

• When the user option was used, it only affected running the daemon.

2.0a3, 2.0a4, 2.0a5 (2007-01-10)

Bugs Fixed

• The new (2.0) mechanism used by zdaemon to start the daemon manager broke some applications that extended zdaemon.

• Added extra checks to deal with programs that extend zdaemon and copy the schema and thus don’t see updates to the ZConfig schema.

2.0a2 (2007-01-10)

New Features

• Added support for setting environment variables in the configuration file. This is useful when zdaemon is used to run programs that need environment variables set (e.g. LD_LIBRARY_PATH).

• Added a command to rotate the transcript log.

2.0a1 (2006-12-21)

Bugs Fixed

• In non-daemon mode, start hung, producing annoying dots when the program exited.

• The start command hung producing annoying dots if the daemon failed to start.

• foreground and start had different semantics because one used os.system and another used os.spawn

New Features

• Documentation

• Command-line arguments can now be supplied to the start and foreground (fg) commands

• zdctl now invokes itself to run zdrun. This means that it’s no-longer necessary to generate a separate zdrun script. This especially when the magic techniques to find and run zdrun using directory sniffing fail to set the path correctly.

• The daemon mode is now enabled by default. To get non-daemon mode, you have to use a configuration file and set daemon to off there. The old -d option is kept for backward compatibility, but is a no-op.

1.4a1 (2005-11-21)

• Fixed a bug in the distribution setup file.

1.4a1 (2005-11-05)

• First semi-formal release.

After some unknown release(???)

• Made ‘zdaemon.zdoptions’ not fail for –help when __main__.__doc__ is None.

After 1.1

• Updated test ‘testRunIgnoresParentSignals’:

o Use ‘mkdtemp’ to create a temporary directory to hold the test socket

rather than creating the test socket in the test directory. Hopefully this will be more robust. Sometimes the test directory has a path so long that the test socket can’t be created.

o Changed management of ‘donothing.sh’. This script is now created by

the test in the temporarily directory with the necessary permissions. This is to avoids possible mangling of permissions leading to spurious test failures. It also avoids management of a file in the source tree, which is a bonus.

• Rearranged source tree to conform to more usual zpkg-based layout:

  o Python package lives under ‘src’.

  o Dependencies added to ‘src’ as ‘svn:externals’.

  o Unit tests can now be run from a checkout.

• Made umask-based test failures due to running as root emit a more forceful warning.

1.1 (2005-06-09)

• SVN tag: svn://svn.zope.org/repos/main/zdaemon/tags/zdaemon-1.1

• Tagged to make better ‘svn:externals’ linkage possible.

To-Dos

More docs:

• Document/demonstrate some important features, such as:

  • working directory

Bugs:

• help command