Merge lp:~sergei.glushchenko/percona-server/5.6-ps-bug1309026 into lp:percona-server/5.6
Proposed by
Sergei Glushchenko
Status: | Merged |
---|---|
Approved by: | Laurynas Biveinis |
Approved revision: | no longer in the source branch. |
Merged at revision: | 612 |
Proposed branch: | lp:~sergei.glushchenko/percona-server/5.6-ps-bug1309026 |
Merge into: | lp:percona-server/5.6 |
Diff against target: |
25 lines (+8/-0) 1 file modified
sql/binlog.cc (+8/-0) |
To merge this branch: | bzr merge lp:~sergei.glushchenko/percona-server/5.6-ps-bug1309026 |
Related bugs: |
Reviewer | Review Type | Date Requested | Status |
---|---|---|---|
Laurynas Biveinis (community) | Approve | ||
Review via email: mp+220879@code.launchpad.net |
Description of the change
Bug 1309026: server crashing, potentially thread pool related
Cause is race condition in group commit code which leads to
concurrent updates in PFS instrumentation while it is not designed
to be updated concurrently. Workaround is to set thread-local
value of PSI_thread to NULL for the follower thread while it is
waiting for leader thread to complete commit.
http://
Debug builds failed because of https:/
To post a comment you must log in.
I wonder about the PFS instrumentation implications of the fix.
Before, PFS tracked the follower thread twice for the same time span: once for the follower thread itself, the waits on m_lock_done and m_cond_done; and at the same time, for the leader thread in the follower context, whatever is done in ha_commit_low.
With the fix, we lose m_lock_done and m_cond_done accounting?