Re: Monitor long running processes and kill them

[Martin Pala]

On Mar 26, 2012, at 9:16 PM, Tom H wrote:

> On 03/26/2012 01:27 PM, Martin Pala wrote:
>> Hi,
>> 
>> if the process does have a pidfile, it is usually updated only when the 
>> process starts - in such case you can use the timestamp test to restart the 
>> process, for example:
>> --8<--
>> check file myprocpid with path /var/run/myproc.pid
>>    if timestamp>  3 days then exec "/etc/init.d/myproc restart"
>> --8<--
>> 
>> 
>> We can easily implement uptime test to Monit in the future, so it can be 
>> possible to use somethig like:
>> --8<--
>> check process myproc with pidfile /var/run/myproc.pid
>>    start program = …
>>    stop program = ...
>>    if uptime>  3 days then restart
>> --8<--
>> 
> 
> Hi,
> 
> These are interesting strategies, as I have some similar requirement.
> 
> For my boxes which are yum based often I see a hung updates yum process that 
> is a few days old, so I am looking for something to go and get those, as they 
> cause a miss on the next run of yum in cron.
> 
> However the outputs of "ps -ef" and "ps aux" are not entirelely trivial to 
> parse, as the "STIME" field seems to rollover from a "00:00" bare time of 
> day, to  a "MMMDD" format after 24 hours...
> 
> Cheers,
> Tom


I have implemented the uptime test and it will be part of the next release … it 
will be possible to use the pattern based process test with the uptime, for 
example:

   check process yum matching "yum"
      if uptime > 1 day then exec "/usr/bin/pkill -9 yum"

=> in the case that process "yum" with uptime of more then 1 day will be found, 
monit will kill it. You can tune/test the pattern on the CLI with "monit 
procmatch <pattern>"

You can get the development snapshot here:
http://www.mmonit.com/tmp/monit-5.4_a1.tar.gz

Regards,
Martin