Bug #4653
closed
mon: creation of cluster with many pgs fails within a reasonble time period
Added by Sage Weil about 11 years ago.
Updated about 11 years ago.
Description
See Jim Schutt's email thread on ceph-devel
- Assignee set to Joao Eduardo Luis
- Category set to Monitor
- Assignee changed from Joao Eduardo Luis to Greg Farnum
- Target version set to v0.61 - Cuttlefish
See Jim's patch on the ML. Need to tune leveldb parameters, but only for monitor, and in the ctor (says sam?).
While we're at it, though, let's add parallel tunables for the OSD, and probably increase the cache size there.
- Status changed from New to In Progress
Preliminary branch in wip-leveldb-config. Going to do some brief tests around it to look at resource consumption.
Initially though I just want to make the minimal set of changes he needed so it would run — we have not in the past enabled any caching whatsoever, and it's not really necessary since it's over and above what the page cache already does.
- Status changed from In Progress to Fix Under Review
Okay, these defaults at least don't balloon memory usage at startup, and they look comparable when doing comparable activities. More specific tests than that will need to be done by somebody else.
Asked for reviews from Joao and SamJ; waiting on those and probably some testing from Jim to merge this.
- Status changed from Fix Under Review to 7
Got reviews from both of them and Sage. Pushed a new version which uses "leveldb" instead of "ldb" in the config names that should be ready for merge (although still waiting on more testing).
- Status changed from 7 to Resolved
Also available in: Atom
PDF