Bug #76373 Problem in UNDO log at file change

• Submitted: 18 Mar 2015 10:29
• Modified: 14 Jan 2016 22:37
• Reporter: Mikael Ronström
• Status: Closed
• Category: MySQL Cluster: Cluster (NDB) storage engine
• Severity: S2 (Serious)
• Version: 7.4.4
• OS: Any
• CPU Architecture: Any

[18 Mar 2015 10:29] Mikael Ronström

Description:
When running disk data tests in autotest we sometimes crash in the middle of restart
processing when we execute the UNDO log.

The problem is related to file changes and the search for end of UNDO log.

We can end up in a situation where we have unwritten pages at the end of the
first UNDO file, then have written parts in the second UNDO file. The
execution of UNDO logs will go backwards and then execute old or even
non-existing log records.

How to repeat:
Run disk data system restart tests for an extended period of time (a couple of hours)
in autotest.

Suggested fix:
Ensure we start executing UNDO log from proper end of log and if started earlier
that we ignore any unwritten pages and faulty pages.

[14 Jan 2016 22:37] Jon Stephens

Fixed in NDB 7.2.21/7.3.10/7.4.7. See BUG#76741 for documentation info.

Closed.