Use FLUSH TABLES before FTWRL
Bug #1277403 reported by
Alexey Kopytov
This bug affects 1 person
| Affects | Status | Importance | Assigned to | Milestone | |
|---|---|---|---|---|---|
| Percona XtraBackup moved to https://jira.percona.com/projects/PXB |
Fix Released
|
Medium
|
Sergei Glushchenko | ||
| 2.1 |
Won't Fix
|
Medium
|
Unassigned | ||
| 2.2 |
Fix Released
|
Medium
|
Sergei Glushchenko | ||
| 2.3 |
Fix Released
|
Medium
|
Sergei Glushchenko | ||
Bug Description
Similar to mysqldump, XtraBackup can potentially lower the impact from FTWRL if an extra FLUSH TABLES is executed before executing FLUSH TABLES WITH READ LOCK. That is, first wait for all currently executing queries to complete without placing a global read lock (and thus, without blocking new queries). Then try to acquire the global read lock.
There is of course always a chance that another long-running query is started in between. If so, we are in the same situation as we were without the extra FLUSH TABLES. Otherwise we get less downtime due to FTWRL.
| tags: | added: low-hanging-fruit |
| description: | updated |
Revision history for this message
| Alexey Kopytov (akopytov) wrote | #1 |
Revision history for this message
| Sergei Glushchenko (sergei.glushchenko) wrote | #2 |
Revision history for this message
| Shahriyar Rzayev (rzayev-sehriyar) wrote | #3 |
To post a comment you must log in.
Would be great to fix this sooner rather than later. It's a one-line fix, and based on some discussions with users I believe there are people who may benefit from it.