Bug#556610: mdadm: Improvement for MD checkarray CRON

[Goswin von Brederlow]

Hi,

do I understand the scenario right that you have a external shared storage
connected to multiple hosts running software raid with a distributed     
filesystem on it?

I would think the right solution would be to pick one dedicated host to run
the monthly check or to assign a month to each node. Your patch to randomize
the day makes it unlikely that 2 nodes run the check on the same day but
you still run the check once per node, producing unneccessary load and wear.

Another solution would be to use a stampfile on the distributed FS. When
the time to check comes lock the file, check the timestamp of the file to
see if it is time to run another check, update the timestamp to signal
a check is under way and run the check. That way only one host would run
the check each month and it should be somewhat random which host wins the lock.

MfG
	Goswin

PS: With the initial idea of running a partial check each night this would
distribute even better.