XtraDB 1.1.8-26.0 assert crash under valgrind

[Sergei Golubchik]

in MariaDB tree, under valgrind, XtraDB crashes like this:

innodb.innodb_prefix_index_liftedlimit 'xtradb' w6 [ fail ]
...
mysqltest: At line 441: query 'DROP TABLE worklog5743_key2' failed: 2013: Lost connection to MySQL server during query
...
120616 6:25:30 InnoDB: Assertion failure in thread 67938640 in file buf0lru.c line 864
InnoDB: Failing assertion: mutex_own(&buf_pool->LRU_list_mutex)

See, for example,
http://buildbot.askmonty.org/buildbot/builders/work-amd64-valgrind/builds/3023/steps/test/logs/stdio

InnoDB does not crash there.

This XtraDB code was introduced in the innodb_split_buf_pool_mutex.patch

[Alexey Kopytov (akopytov)]

It's a regression introduced in 5.5 when merging the upstream fix for http://bugs.mysql.com/bug.php?id=64284. The new code in buf_flush_dirty_pages() locks the global buffer pool mutex rather than LRU_list_mutex.

This specific bug only applies to debug builds, because buf_LRU_insert_zip_clean() a UNIV_DEBUG-only function. But we should inspect all recently added calls to buf_pool_mutex_enter() in general and the fix for #51325 in particular to verify if they are compatible with innodb_split_buf_pool_mutex.patch.

[Alexey Kopytov (akopytov)]

Same problem in a Jenkins debug (but not Valgrind) build: http://jenkins.percona.com/view/PS%205.5/job/percona-server-5.5-param/437/BUILD_TYPE=debug,Host=ubuntu-oneiric-32bit/testReport/junit/%28root%29/innodb/innodb_prefix_index_liftedlimit/

[Shahriyar Rzayev (rzayev-sehriyar)]

Percona now uses JIRA for bug reports so this bug report is migrated to: https://jira.percona.com/browse/PS-567