terrcin / niet Goto Github PK

The only change I've made is to send a QUIT instead of a TERM when receiving a QUIT so as to interact 
with Resque (https://github.com/defunkt/resque) better. Resque expects a QUIT to graceful shutdown 
workers which are actively processing a task rather than a TERM.
====================================================================================================


Niet, A process respawner for people who want syslogging and very little else
=============================================================================


Usage: niet /usr/bin/someprogram foo bar
          - Runs someprogram with the arguments 'foo' and 'bar', restarting the program again if it
            dies, waiting for up to 60s if it's dying in less than 60s. Sends output from the program
            on its stdout and stderr to syslog using `logger`.  If sent a TERM signal, sends a TERM
            signal to the program, waits for it to finish, and then restarts it.  If sent a QUIT
            signal, sends a QUIT signal to the program, waits for it to finish, and then quits.

            niet requires no privileges and should be run as the user you want to run the daemon under.
            It can be run as root, but running your daemons as root is generally discouraged.

Options: -o daemon.notice  Changes the log priority of the syslog messages logged from the program's
                           stdout to 'daemon.notice'.  Default: user.info.
         -e daemon.alert   Changes the log priority of the syslog messages logged from the program's
                           stderr to 'daemon.alert'.  Default: user.err.
         -t syslog_tag     Changes the syslog tag of the syslog messages logged from the program's
                           stdout & stderr to 'syslog_tag'.  Default: the program's command name.
         -k 15             Sets a timeout of 15 seconds after the program is sent a QUIT or TERM 
                           signal, after which it will be killed by a KILL signal.  Default: no KILL 
                           signal.
         -p /var/run/x.pid Writes the PID of this process to /var/run/x.pid.  Default: no PID file.
                           The use of PID files is discouraged: they're just another thing to go
                           wrong.  niet is designed to be controlled entirely using signals, which
                           you can use without PID files, eg. `killall niet` or `killall -QUIT niet`;
                           you can tell when niet (and therefore the supervised program) has
                           terminated because there's nothing left to kill.
         -c /cd/to/here    Changes to this directory before running the command.  Default: /.


Installation
============

To compile niet and install it to /usr/local/bin (which is usually already on your PATH), run:

	make && sudo make install


Deploying application daemons with niet
=======================================

See DEPLOYING.


Non-features
============

Niet does not:

 - need to run as root

 - make or listen on sockets or IPC channels of any kind

 - have a complex and buggy process lifecycle

 - fill your logs and use your CPU spinning around restarting the program rapidly if it's repeatedly
   failing (it'll wait (60s - the time the last run survived for), eg. 50s if the program lasted 10s;
   so it won't wait if the program survived at least a minute, and it won't wait when you signal it
   to restart the program)

 - give up starting the program (datacentre clusters frequently take >30 minutes to become fully
   operational if the database server fscks when the power comes back on)

 - mess about with logfiles (niet is for people with multiple servers, who like syslog -
   try supervisord if you want logfiles).

 - send you notifications (you should have alerting on your syslogs so you can see errors from
   everything); niet will log all 'business as usual' messages, including requested restarts, at the
   'normal' priority, and all 'not business as usual' messages, including unexpected app crashes and
   any app output on STDERR, at the 'error' priority; you can customize these using the above options.

 - poll and restart your program if it's using too much memory (your operating system can
   automatically limit a program's memory use, without polling - use ulimit)

 - poll and restart your program if it's using too much CPU (you want your app to crash when there's
   a lot of work on?  use something else, also learn to code)

 - monitor the state of your server (niet is for processes - try munin)


License
=======

MIT.