Fun with Bugs #71 - On Some Public Bugs Fixed in MySQL 5.7.24

Oracle released many new MySQL versions back on Monday, but I had no time during this very busy week to check anything related (besides the fact that MySQL 8.0.13 can be complied from source on my Fedora 27 box). I am sure you've read a lot about MySQL 8.0.13 elsewhere already, even patches contributed by Community are already presented in a separate post by Jesper Krogh.

I am still mostly interested in MySQL 5.7. So, here is my typical quick review of some selected bugs reported in public by MySQL Community users and fixed in MySQL 5.7.24.

My wife noticed this nice spider in the garden and reported it to me via this photo. Spider is formally not a bug, while in this post I discuss pure bugs...

Let me start with fixes in Performance Schema (that is supposed to be mostly bugs free):

• Bug #90264 - "Some file operations in mf_iocache2.c are not instrumented". This bug reported by Yura Sorokin from Percona, who also contributed patches, is fixed in all recent Oracle releases, from 5.5.62 to 8.0.13.
• Bug #77519 - "Reported location of Innodb Merge Temp File is wrong". This bug was reported by Daniël van Eeden back in 2015. Let's hope files are properly reported in @@tmpdir now.
• Bug #80777 - "order by on LAST_SEEN_TRANSACTION results in empty set". Yet another bug report from Daniël van Eeden got fixed.

Let's continue with InnoDB bugs:

• Bug #91032 - "InnoDB 5.7 Primary key scan lack data". Really weird bug was reported by Raolh Rao back in May.
• Bug #95045 - Release notes are referring to public bug that does not exist! So, we have a bug in them. Related text:
  

  "It was possible to perform FLUSH TABLES FOR EXPORT on a partitioned table created with innodb_file_per_table=1 after discarding its tablespace. Attempting to do so now raises ER_TABLESPACE_DISCARDED. (Bug #95045, Bug #27903881)"

  and refer to Bug #80669 - "Failing assert: fil_space_get(table->space) != __null in row0quiesce.cc line 724", reported by Ramesh Sivaraman from Percona. In the comment from Roel there we see that actual bug was Bug #90545 that is, surprise, still private!
  Recently I found out (here) that some community members think that keeping crashing bugs private after the fixed version is released is still better than publish test cases for them before all affected versions are fixed... I am not so sure.

What about replication (group replication aside, I have enough Galera problems to deal with in my life to even think about it)? There are some interesting bug fixes:

• Bug #90551 - "[MySQL 8.0 GA Debug Build] Assertion `!thd->has_gtid_consistency_violation'". Good to know that Oracle engineers still pay attention to debug assertions, as in this report (with nice simple test case involving XA transactions)  reported by Roel Van de Paar from Percona.
• Bug #89370 - "semi-sync replication doesn't work for minutes after restart replication". This bug was reported by Yan Huang, who had contributed a patch for it.
• Bug #89143 - "Commit order deadlock + retry logic is not considering trx error cases". Nice bug report from Jean-François Gagné.
• Bug #83232 - "replication breaks after bug #74145 happens in master". FLUSH SLOW LOGS that failed on master (because of file permission problem, for example) was still written to the binary log. Nice finding by Jericho Rivera from Percona.

There are interesting bugs fixed in other categories as well. For example:

• Bug #91914 - "Mysql 5.7.23 cmake fail with 'Unknown CMake command "ADD_COMPILE_FLAGS".'" Now thanks to this report by Tomasz Kłoczko one can build MySQL 5.7 with gcc 8.
• Bug #91080 - "sql_safe_updates behaves inconsistently between delete and select". The fix is described as follows:
  

  "For DELETE and UPDATE that produced an error due to sql_safe_updates being enabled, the error message was insufficiently informative. The message now indicates that data truncation occurred or the range_optimizer_max_mem_size value was exceeded.
  
  Additionally: (1) Using EXPLAIN for such statements does not produce an error, enabling users to see from EXPLAIN output why an index is not used; (2) For multiple-table deletes and updates, an error is produced with safe updates enabled only if the target table or tables use a table scan."

  I am NOT sure this is the fix that bug reporter, Nakoa Mccullough, was expecting. He asked to be consistent with SELECT (that works). The bug is still closed :(
• Bug #90624 - "Restore dump created with 5.7.22 on 8.0.11". It seems Emmanuel CARVIN asked for the working way to upgrade from 5.7.x to 8.0.x. Last comment seems to state that upgrade from 5.7.24 to 8.0.13 is still not possible. I had not checked this.
• Bug #90505 is private. Release notes say:
  

  "If flushing the error log failed due to a file permission error, the flush operation did not complete. (Bug #27891472, Bug #90505) References: This issue is a regression of: Bug #26447825"

  OK, we have a private regression bug, fixed. Nice.
• Bug #90266 - "No warning when truncating a string with data loss". It was when making BLOB/TEXT columns smaller. Nice finding by Carlos Tutte.
• Bug #89537 - "Regression in FEDERATED storage engine after GCC 7 fixes". Yet another by report with patch contributed by Yura Sorokin.
• Bug #88670 - "Subquery incorrectly shows duplicate values on subqueries.". Simple from results bug in optimizer affecting all versions starting from 5.6. Fixed now thanks to Mark El-Wakil.

That's all bugs I wanted to mention today. To summarize my feelings after reading the release notes:

1. I'd surely consider upgrade to 5.7.24 in any environment where replication is used. Some InnoDB fixes also matter.
2. We still see not only private bugs (with questionable security impact) mentioned in the release notes, but this time also a typo in bug number that makes it harder to find out what was really fixed and why.
3. I think it would be fair for Oracle to mention Percona as a major contributor to MySQL 5.7, in a same way as Facebook is mentioned in many places with regards to 8.0.13.
4. It's good to know that some debug assertions related bugs are still fixed. More on this later...