Bug #15665: Live journal replay might skip entries from previous object set - rbd - Ceph

Actions

Copy link

Bug #15665

closed

Live journal replay might skip entries from previous object set

Added by Jason Dillaman almost 8 years ago. Updated almost 8 years ago.

Status:

Resolved

Priority:

High

Assignee:

Jason Dillaman

Target version:

% Done:

Source:

other

Tags:

Backport:

jewel

Regression:

Severity:

3 - minor

Reviewed:

Affected Versions:

ceph-qa-suite:

Pull request ID:

Crash signature (v1):

Crash signature (v2):

Description

The last entry from a previously polled version of an object resulted in the object being trimmed from the replay set:

2016-04-29 11:46:06.056047 7f6c877fe700 20 JournalPlayer: try_pop_front
2016-04-29 11:46:06.056068 7f6c877fe700 20 JournalPlayer: advance_splay_object: new offset 3
2016-04-29 11:46:06.056071 7f6c877fe700 15 JournalPlayer: journal_data.1.102d74b0dc51.2 empty
2016-04-29 11:46:06.056074 7f6c877fe700 10 JournalPlayer: fetch: journal_data.1.102d74b0dc51.6
2016-04-29 11:46:06.056086 7f6c877fe700 10 ObjectPlayer: fetch: journal_data.1.102d74b0dc51.6
2016-04-29 11:46:06.056162 7f6c877fe700 20 JournalMetadata: allocated commit tid: commit_tid=15311 [object_num=2, tag_tid=2, entry_tid=15306]

This caused the live replay to start polling the object from the next set:

2016-04-29 11:46:07.125631 7f6d01409700 10 ObjectPlayer: handle_watch_task: journal_data.1.102d74b0dc51.6 polling
2016-04-29 11:46:07.125634 7f6d01409700 10 ObjectPlayer: fetch: journal_data.1.102d74b0dc51.6
2016-04-29 11:46:07.145842 7f6cca7fc700 10 ObjectPlayer: handle_fetch_complete: journal_data.1.102d74b0dc51.6, r=0, len=598896

In this case, we shouldn't remove an object from the replay until we poll it for one last time.

Related issues 1 (0 open — 1 closed)