Bug #33282 | broken consistency when read-committed or innodb_locks_unsafe_for_binlog | ||
---|---|---|---|
Submitted: | 17 Dec 2007 6:00 | Modified: | 17 Dec 2007 16:42 |
Reporter: | Yasufumi Kinoshita | Email Updates: | |
Status: | Duplicate | Impact on me: | |
Category: | MySQL Server | Severity: | S1 (Critical) |
Version: | 5.1.22-rc | OS: | Any |
Assigned to: | CPU Architecture: | Any |
[17 Dec 2007 6:00]
Yasufumi Kinoshita
[17 Dec 2007 11:03]
Domas Mituzas
Seems to be reincarnation of Bug#31310
[17 Dec 2007 13:50]
Heikki Tuuri
Yasufumi, thank you again for a very good bug report! Trx 2 should return with a 'semi-consistent read' to MySQL the latest committed state of the row, which is (1, 0). MySQL sees that the UPDATE statement SHOULD update that. MySQL should reissue the UPDATE for that row, this time doing a normal locking read. That is, trx 2 should wait for trx 1 to commit. This may be a bug in MySQL's SQL interpreter like Bug#31310. MySQL probably fails to reissue the UPDATE for the row.
[17 Dec 2007 16:42]
Susanne Ebrecht
Many thanks for writing a bug report. This is a duplicate of bug #31310.