Actions
Bug #1564
closedosd: osd should not be primary before data is replicated
Status:
Won't Fix
Priority:
Normal
Assignee:
-
Category:
OSD
Target version:
-
% Done:
0%
Source:
Tags:
Backport:
Regression:
Severity:
Reviewed:
Affected Versions:
ceph-qa-suite:
Pull request ID:
Crash signature (v1):
Crash signature (v2):
Description
Right now an OSD is kicked out of the acting set only if it needs a backlog, and it's allowed back in to the acting set once it has the backlog. This means that it can become primary for a PG while it still has hours of data to transfer, though! This is pretty bad since any reads or writes to that PG are likely to block...
Now, the OSD is supposed to prioritize objects which have waiting operations, but that's still a pretty significant latency increase, and sometimes it breaks (see #1563).
Updated by Sage Weil over 12 years ago
- Translation missing: en.field_position set to 28
Updated by Sage Weil over 12 years ago
- Target version deleted (
v0.37) - Translation missing: en.field_position deleted (
28) - Translation missing: en.field_position set to 11
Updated by Sage Weil over 12 years ago
- Subject changed from OSD should not be primary before data is replicated to osd: osd should not be primary before data is replicated
- Translation missing: en.field_position deleted (
23) - Translation missing: en.field_position set to 23
Updated by Sage Weil over 12 years ago
- Translation missing: en.field_position deleted (
23) - Translation missing: en.field_position set to 112
Updated by Sage Weil about 12 years ago
- Status changed from New to Won't Fix
no more backlogs, so this problem is mostly moot. it can sort of still happen (to a vastly decreased degree), but it's not immediately clear that it is worth the cost of remapping the pg.
Actions