I have an agent which monitors a process and restarts it.
It works fine.
The one thing I would like to improve is how quickly it notices that a process
has died and fails over or restarts it.
I know that I can increase the monitoring level to do this. However, it seems
like a lot of wasted cycles. Instead of polling I would really like a callback.
I searched the archives for "process death" and did not see any comments in
this area.
I think the basic requirements are:
1.When a process starts it registers itself with a kernel component. This
registration also gets passed an action.
2.If the process is gracefully exiting, it unregisters itself.
3.If the process dies, the kernel component would call the action.
I assume this is already there but I have not found it thus far. It is similar
to the watchdog but I do not want to kill the whole box, I just want to trigger
Pacemaker to do the monitor/start cycle.
A possible implementation is:
The registration opens /dev/process_death and leaves it open
Graceful exit does an ioctl to /dev/process_death that going down gracefully.
If a close on /dev/process_death occurs, we know that the process died and we
need to call an action. The action would then somehow trigger Pacemaker to do
the monitor/start cycle of the resource.
Is something like this already there or something better?
Thanks,
Bob
_______________________________________________________
Linux-HA-Dev: [email protected]
http://lists.linux-ha.org/mailman/listinfo/linux-ha-dev
Home Page: http://linux-ha.org/
