Bug #1898
closed
very long scrub blocked write operation
Added by Greg Farnum over 12 years ago.
Updated over 12 years ago.
Description
On cephstore6235 we saw a write operation get blocked for 4 minutes by scrub. The log is available in /var/log/ceph/long_scrub.log, operation client.898189.0:13131563
- Translation missing: en.field_position set to 3
- Target version deleted (
v0.41)
- Translation missing: en.field_position deleted (
10)
- Translation missing: en.field_position set to 5
- Priority changed from Immediate to Urgent
- Target version set to v0.41
It seems that osd.220 took 5 minutes between starting _scan_list and relocking the pg.
2012-01-06 16:22:52.45141 another case where osd.220 required 4 minutes to scan the 150000 object pg. osd.97 seems to require closer to 2 minutes to scan the pg. We can reduce the severity of this problem by waiting for all replicas to respond to the first scan before stopping writes.
- Status changed from New to Won't Fix
We will fix this as part of the incremental scrubbing #1783
- Status changed from Won't Fix to Duplicate
- Priority changed from Urgent to High
Even though this is not the same complaint as 1783, we plan on
fixing it with the same changes, so I am calling this a DUP of
1783.
Also available in: Atom
PDF