Friday, July 21, 2017

Fun with Bugs #54 - On Some Bugs Fixed in MySQL 5.7.19

More than 3 months after 5.7.18 we' ve got MySQL 5.7.19 released recently. This is my quick review of the release notes with interesting fixed bug (reported in public) highlighted in the areas I am usually interested in.

Let's start with InnoDB. The following bug fixes attracted my attention:
  • Bug #85043 is still private. You know how much I hate those. At least we can see it was about the fact that "The server allocated memory unnecessarily for an operation that rebuilt the table." Let's hope this is no longer the case.
  • Bug #84038 - "Errors when restarting MySQL after FLUSH TABLES FOR EXPORT, RENAME and DROP.", was reported by Jean-François Gagné and verified by Umesh Shastry. Basically, InnoDB failed to update INNODB_SYS_DATAFILES internal data dictionary table properly while renaming tables.
  • Bug #80788 - "Reduce the time of looking for MLOG_CHECKPOINT during crash recovery". It was reported by  Zhai Weixiang (who had provided a patch) and quickly verified by Umesh Shastry.
  • Bug #83470 - "Reverse scan on a partitioned table does ICP check incorrectly, causing slowdown". This report that is related to both InnoDB, partitioning and optimizer, was created by my colleague from MariaDB, Sergey Petrunya, who had also suggested a patch under BSD license. Bug #84107 was considered a duplicate of this one it seems.
There were some public bugs fixed in group replication. While I do not really care, yet, about Group Replication, it's nice to see bugs in this area fixed so fast:
  • Bug #85667 - "GR node is in RECOVERING state if binlog_checksum configured on running server". This bug was reported by Ramana Yeruva.
  • Bug #85047 - "Secondaries allows transactions through ASYNC setup into the group". Yet another group replication bug recently fixed. It was reported by Narendra Chauhan who probably works for Oracle.
  • Bug #84728 - "Not Possible To Avoid MySQL From Starting When GR Cannot Start", was reported by Kenny Gryp from Percona.
  • Bug #84329 - "Some Class B private address is not permitted automatically". This funny bug was reported by Satoshi Mitani and verified by Umesh Shastry
  • Bug #84153 - "ASSERT "!contains_gtid(gtid)" Owned_gtids::add_gtid_owner". It was reported by Narendra Chauhan.
Usual async replication was also improved in 5.7.19, with the following related bugs fixed:
  • Bug #83184 - "Can't set slave_skip_errors > 3000". This bug was reported by Tsubasa Tanaka (who had provided a patch) and verified by Umesh Shastry.
  • Bug #82283 - "main.mysqlbinlog_debug fails with a LeakSanitizer error". Laurynas Biveinis from Percona found it and provided patches. The bug was verified by Umesh Shastry.
  • Bug #81232 - "Changing master_delay after stop slave results in loss of events". Ths regression bug (5.6.x was not affected) was reported by Parveez Baig. Bug #84249 (reported by Sergio Roysen) was declared a duplicate of this one.
  • Bug #77406 - "MTS must be able handle large packets with small slave_pending_jobs_size_max", was reported by my colleague Andrii Nikitin while he worked in Oracle.
  • Bug #84471 is still private. Why it could be so when the bug is supposed to be about "...the master could write to the binary log a last_committed value which was smaller than it should have been. This could cause the slave to execute in parallel transactions which should not have been, leading to inconsistencies or other errors." Let's hide the details about all bugs that may lead to data corruption,s why not?
  • Bug #83186 - "Request for slave_skip_errors = ER_INCONSISTENT_ERROR". Thanks to  Tsubasa Tanaka, who had provided a patch, we can do this now.
  • Bug #82848 - "Restarting a slave after seeding it with a mysqldump loses it's position". This serious problem with GTID-based replication was noted and resolved (with a patch) by DaniĆ«l van Eeden.
  • Bug #82209 is also private... Now, I am tired of them, just go real the release notes and hope they describe the problem and solution right. Great for any open source software, isn't it?
  • Bug #81119 - "multi source replication slave sql running error 1778". This serious bug (where GTIDs also play role) was reported by Mohamed Atef. Good to see it fixed finally.
I know long term query cache is doomed, Oracle is not going to work on it any more, but in the meantime, please, check this important bug fixed, Bug #86047 - "FROM sub query with 'group by' is cached by mistake", by Meiji Kimura. If you still use query cache and have complex queries with derived tables, please, consider upgrade. You might be getting wrong results all this time...

If you use xtrabackup or Oracle's MySQL Enterprise Backup in the environment with XA transactions, please, check Bug #84442 - "XA PREPARE inconsistent with XTRABACKUP", by David Zhao, who had also provided a patch. Your backups may be inconsistent until you upgrade...

Now, on some optimizer bugs fixed:
  • Bug #81031 - "count(*) on innodb sometimes returns 0". This happened when index merge was used by the optimizer. The bug was reported by Ashraf Amayreh and verified by Shane Bester.
  • Bug #84320 - "DISTINCT clause does not work in GROUP_CONCAT". It was reported by Varun Gupta and verified by Bogdan Kecman. See Bug #68145 also (that is still "Verified").
  • Bug #79675 - "index_merge_intersection optimization causes wrong query results". This bug was reported by Saverio M and verified by Miguel Solorzano.
There were a lot more bugs fixed in 5.7.19. I'd say that if one uses replication (especially GTID-based or group replication), upgrade is a must, but I had no tried this version myself, yet.

No comments:

Post a Comment