Tuesday, April 10, 2018

Fun with Bugs #65 - On MySQL Bug Reports I am Subscribed to, Part V

I think it's time to review some bugs I've subscribed to several months ago, those older than in the first post from this series. There are several really serious bugs in the list of 15 below:
  • Bug #87560 - "XA PREPARE log order error in replication and binlog recovery". This bug was reported by Wei Zhao, who also provided patches.
  • Bug #87526 - "The output of 'XA recover convert xid' is not useful". This bug reported by Sveta Smirnova is well known and is a real pain for DBAs who have to deal with incomplete XA transactions after some crash or unexpected restart. Check PS-1818 and MariaDB task MDEV-14593. The problem is resolved in MariaDB 10.3.3+ by a new XA RECOVER FORMAT='SQL' option.
  • Bug #87164 - "Queries running much slower in version 5.7 versus 5.6". It was reported by  Alok Pathak from Percona and stays "Verified" since August, 2017.
  • Bug #87130 - "XA COMMIT not taken as transaction boundary". Yet another XA bug report with a patch contributed by Wei Zhao.
  • Bug #87084 - "FK DELETE CASCADE does not honor innodb_lock_wait_timeout". Nice report by Elena Stepanova from MariaDB. As you can find out from MDEV-15219, it's properly fixed in MariaDB 10.2.13+.
  • Bug #87065 - "Release lock on table statistics after query plan created". Great feature request by Sveta Smirnova. The actual problem behind this feature request was resolved by Percona in versions 5.7.20-18+. This fix is one of few really good reasons to use recent Percona Server 5.7, so I opened MDEV-15101 for MariaDB also.
  • Bug #86926 - "The field table_name (varchar(64)) from mysql.innodb_table_stats can overflow.". I really wonder why this bug report by Jean-François Gagné still remains just "Verified".
  • Bug #86865 - "InnoDB does unnecessary work when extending a tablespace". Bug report and patch by Alexey Kopytov.
  • Bug #86705 - "Memory leak of Innodb". Great example of a leak (in MySQL 5.5.x only) found with a help of Valgrind/Massif. Qinglin Zhang also suggested a simple patch.
  • Bug #86475 - "Error with functions and group by with ONLY_FULL_GROUP_BY". Nice feature request from Arnaud Adant.
  • Bug #86462 - "mysql_ugprade: improve handling of upgrade errors". Simon Mudd asked for some better error messages at least, so that running under strace would not be needed.
  • Bug #86215 - "MySQL is much slower in 5.7 vs 5.6". This report from Mark Callaghan includes a lot of results and details on performance regressions at low concurrency starting from 5.0 and up to 8.0. For some cases studied the biggest drop in QPS is from 5.6 to 5.7.
  • Bug #86163 - "can't update temporary table when joined with table with triggers on read-only". I'd call this bug found by Bret Westenskow funny. The rest of them in this list are serious.
  • Bug #85970 - "Memory leak with transactions greater than 10% of the total redo log size". Nice corner case was studied by Joffrey MICHA├ĆE. Whenever you see
    The size of BLOB/TEXT data inserted in one transaction is greater than 10% of redo log size...
    error messages, take care, as memory allocated may not be released.
  • Bug #85910 - "Increased Performance_Schema overhead on Sending data". Seems to be a known problem fixed in MySQL 5.7. Still a nice report by Jervin R on the overhead one may expect from MySQL 5.6 on some systems.
So, we traveled back in time for a year. I am really sorry to see XA-related bugs and some reports with patches properly contributed still active.


No comments:

Post a Comment