===============
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>
.. -> text
>>> with open('conf', 'w') as file:
... _ = file.write(text)
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>
.. -> text
>>> with open('conf', 'w') as file:
... _ = file.write(text.replace('/tmp', tmpdir))
Now, when we run zdaemon:
sh> ./zdaemon -Cconf start
. .
daemon process started, pid=1139
sh> ls
conf
zdaemon
.. test
>>> import os
>>> os.path.exists("/tmp/demo.zdsock".replace('/tmp', tmpdir))
True
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>
.. -> text
>>> with open('conf', 'w') as file:
... _ = file.write(text.replace('/tmp', tmpdir))
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>
LD_LIBRARY_PATH /home/foo/lib
HOME /home/foo
</environment>
.. -> text
>>> with open('conf', 'w') as file:
... _ = file.write(text.replace('/tmp', tmpdir))
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
LD_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>
.. -> text
>>> with open('conf', 'w') as file:
... _ = file.write(text)
Now we'll start:
sh> ./zdaemon -Cconf start
. .
daemon process started, pid=7963
.. Wait a little bit to make sure tail has a chance to work
>>> import time
>>> time.sleep(0.1)
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())
.. Wait a little bit to make sure tail has a chance to work
>>> time.sleep(1)
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())
.. Wait a little bit to make sure tail has a chance to work
>>> time.sleep(1)
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.
