Bug #9987
closed
mon: min_last_epoch_complete tracking broken
Added by Sage Weil over 9 years ago.
Updated over 8 years ago.
Description
When we moved to pulling pgmap values out of keys we broke the min_last_epoch_clean invalidation code.
I suspect this is why people see leveldb grow... and a mon restart fixes it?
BTW, we compact our dumpling mon leveldb without restarting. We do
ceph tell mon.0 compact
and that can shrink the db from 25GB down to ~1GB. Maybe it would compact even further with this fix?
I don't think this will help much with your case. This patch will allow the monitor to delete data that should be removed from the store. Compaction will only work to remove already deleted data. It may certainly help with disk usage footprint by reducing the amount of data the monitors keep at any given time, and compaction will certainly help blowing those extra MBs away.
However, I've been wondering if keeping this data around (especially if it's lots of data) hinders leveldb automated compaction. If that's true, then I think this patch could indeed help out a lot with your issues, and I would be happy to buy a microbrewery for sage so that he'd have a lifetime of free beer.
- Status changed from Fix Under Review to Pending Backport
- Status changed from Pending Backport to Resolved
- Backport changed from giant, firefly, dumpling to giant, firefly
Also available in: Atom
PDF