Tuesday, January 16, 2018

Fun with Bugs #60 - On Some Memory Leaks, Replication and Other Bugs Fixed in MySQL 5.7.21

Oracle had formally released MySQL 5.7.21 yesterday. I do not bother any more to study MySQL release notes carefully and completely, but during a quick review today I've noted several interesting items I'd like you to pay attention to.

I am historically interested in InnoDB implementation details, so I could not miss Bug #87619 - "InnoDB partition table will lock into the near record as a condition in the use ". This was a regression bug in 5.7+, probably caused by new implementation of partitioning in InnoDB.

Another interesting bug is Bug #86927 - "Renaming a partitioned table does not update mysql.innodb_table_stats.", by Jean-François Gagné. It was yet another bug in InnoDB's persistent statistics (that I truly hate). What makes it especially interesting to me, though, is that it's the first public bug report I noted that mentioned MySQL 9.0.0 release as a target for the fix:
"Fixed as of the upcoming 5.7.21, 8.0.4, 9.0.0 release"
So, it's clear that back in October 2017 Oracle had already got a separate branch for upcoming MySQL 9.0.x! It also probably means that MySQL 8.0.x GA is coming really soon.

There are bug reports that are worth reading for technical reasons, others - only if you want to get some fun. Bug #86607 - "InnoDB crashed when master thread evict dict_table_t object" is agood example that covers both cases. Good to know the crash is fixed, but, please, make sure to read all comments there.

In this release I've noted fixes to several public bugs reported by Shane Bester. The first one of them is Bug #86573 - "foreign key cascades use excessive memory". Check how he used memory instrumentation in Performance Schema to demonstrate the problem! In Bug #86482 - "innodb leaks memory, performance_schema file_instances #sql-ib3129987-252773.ibd", he used similar approach to show potential memory leak in the Performance Schema itself ! Yet another bug that mentions 9.0.0 as a target version for the fix, among others... 

Bug #78048 - "INNODB Full text Case sensitive not working", is here both because I recently started to notice problems related to InnoDB FULLTEXT indexing, again (first time was soon after it was introduced), and because it has an MTR  test case contributed by Sveta Smirnova.


XA transactions support had always been problematic in MySQL  (still "Verified" Bug #87526 by Sveta Smirnova is one of recent examples how incomplete or useless it can be, see also MDEV-14593). Check the following bugs fixed in MySQL 5.7.21 if you use XA transactions:
  • Bug #87393 - "xa rollback with wrong xid will be recorded into the binlog". It was reported by HongXiang Jiang, who had also contributed a patch.
  • Bug #83295 - "replication error occurs, use xa transaction(one phase)". Yet another XA transactions problem reported by Hiroyuki Itoh and then confirmed by many affected users. Nice to see it fixed.
There are many fixes in MySQL 5.7.21 related to memory leaks. Two bug reports of this kind were from Przemyslaw Malkowski:
  • Bug #85371 - "Memory leak in multi-source replication when binlog_rows_query_log_events=1". Again, memory instrumentation of Performance Schema was used to demonstrate the problem. Vlad Lesin, also from Percona, contributed the patch for this bug.
  • Bug #85251 - "Memory leak in master-master GTID replication with sync_relay_log_info". Here Vlad Lesin, who had contributed the patch, also used Massif for the detailed analysis.
To summarize, I start to miss memory instrumentation in Performance Schema in MariaDB 10.x... This is a really useful feature.

I usually care about optimizer bugs, and these two attracted my attention:
  • Bug #87207 - "select distinct with secondary key for 'Using index for group-by' bad results". This nice optimizer regression bug was found by Shane Bester. As a workaround, while you do not use 5.7.21, you can try to set optimizer_switch='use_index_extensions=off'. I'd keep it that way by default...
  • Bug #72854 - "Extremely slow performance with outer joins and join buffer". I am happy to see this old optimizer bug reported by Sergey Petrunya from MariaDB finally fixed.
You can find a lot more details, including usual references to MySQL bug reports that are still private, in the Release Notes. Keep reading and consider upgrade :)

No comments:

Post a Comment