Bug #250
closed
mon: delete old states to avoid filling disk
Added by Sage Weil almost 14 years ago.
Updated over 12 years ago.
Description
We trim the pgmap (which gets the most activity), but not the others: logs, mdsmap, osdmap, etc. Unfortunately the others are trickier. mds may be safe, but for the osdmap the clients/osds currently needs a full history from where they are to present in order to behave. Need to be careful about throwing out old states. Maybe a min age would help?
I keep running into this, my monitors keep running out of inodes which causes them to stall.
Right now I have cron's like:
@daily find /var/lib/ceph/mon.logger/logm -type f -ctime +7 -delete
pgmap already gets trimmed, but the logm doesn't.
I was thinking about a cron which runs every day and ships with the Ceph packages, would that be OK? Or do you prefer to implement this in the monitor itself?
We should probably just do this in the monitor. The logm is never read so it's safe to delete whenever, unlike the other maps we're concerned about.
- Target version deleted (
12)
- Target version set to v0.38
- Translation missing: en.field_position deleted (
569)
- Translation missing: en.field_position set to 958
- Status changed from New to Resolved
- Translation missing: en.field_position deleted (
958)
- Translation missing: en.field_position set to 958
- Translation missing: en.field_position deleted (
958)
- Translation missing: en.field_position set to 958
Also available in: Atom
PDF