Sunday, April 1, 2018

Fun with Bugs #64 - On MySQL Bug Reports I am Subscribed to, Part IV

I've subscribed to more than 15 new MySQL bug reports since the previous post in this series, so it's time for a new one. I am trying to follow important, funny or hard to process bug reports every day. Here is the list of the most interesting recent ones starting from the latest (with several still not processed properly):
  • Bug #90211 - "Various warnings and errors when compiling MySQL 8 with Clang".  Roel Van de Paar and Percona in general continue their QA efforts in a hope to make MySQL 8 better. Current opinion of Oracle engineers on this bug is the following:
    "First of all, these issues are in protobuf, not MySQL per se. There are some warnings with Clang 6, but since they're in third-party code, we have simply disabled them when compiling protobuf (will be part of 8.0.11). Optionally, -DUSE_SYSTEM_LIBS=1 will use system protobuf and thus not compile the files in question.
    As for the crash, we don't support prerelease compilers (more generally, we support platforms, not compilers). Given the stack trace, it is highly likely that the issue either is in the prerelease Clang, or in protobuf.
    Let's see how it may end up. Roel rarely gives up easily...
  • Bug #90209 - "Performance regression with > 15K tables in MySQL 8.0 (with general tablespaces)". Nice regression bug report from Alexander Rubin. It is still "Open".
  • Bug #90190 - "Sig=6 assertion in MYSQL_BIN_LOG::new_file_impl |". Yet another bug report from Percona employee, Ramesh Sivaraman.
  • Bug #89994 - "INDEX DIRECTORY shown as valid option for InnoDB table creation". Everybody knows how much I like fine MySQL manual. Even more I like when missing or wrong details are found there, like in this case reported by by colleague from MariaDB, Claudio Nanni.
  • Bug #89963  - "Slowdown in creating new SSL connection". Maybe it's comparing apples to oranges, as stated in one of comments, but I am surprised that this (performance regression) bug report by Rene' Cannao' is still "Open". It requires more attention, IMHO. Speed of connections matters a lot for MySQL.
  • Bug #89904 - "Can't change innodb_max_dirty_pages_pct to 0 to flush all pages". Good intentions to set better default value (applied a bit later than needed) led to the problem. As Simon Mudd put it:
    "innodb_max_dirty_pages_pct_lwm setting has existed since 5.6. This issue only comes up as by changing the default value to 10 those of us who have ignored it until now never noticed it existed. That is a shame as setting this value to a value other than 0 (e.g. 10 which is the new default) should be better and trigger some background flushing of dirty pages avoiding us hitting innodb_max_dirty_pages_pct which would trigger much more aggressive behaviour which is not really desirable."
  • Bug #89876 - "mysqladmin flush-hosts is not safe in GTID mode". Yet another bug report from Simon Mudd. See also Bug #88720 that highlights even more problems with various FLUSH statements and GTIDs.
  • Bug #89870 - "Group by optimization not used with partitioned tables". For some reason this report from Arnaud Adant is still "Open". As my colleague Richard Stracke stated:
    "The only solution would be, that the optimizer is able to check, if the condition in the where clause include the whole table (or partition) and in this case use group by optimization."
  • Bug #89860 - "XA may lost prepared transaction and cause different between master and slave." As this (and other, like Bug #88534) bug report from Michael Yang shows, there is still a long way to go until it would be safe to use XA transactions with MySQL.
  • Bug #89834 - "Replication will not connect on IPv6 - does not function in an IPv6 only environ". This bug report from Tim St. Pierre is still "Open".
  • Bug #89822 - "InnoDB retries open on EINTR error only if innodb_use_native_aio is enabled". We have patch contributed by Laurynas Biveinis from Percona.
  • Bug #89758 - "Conversion from ENUM to VARCHAR fails because mysql adds prefix index". This funny bug was found and reported by Monty Solomon.
  • Bug #89741 - "Events log Note level messages even for log_warnings=0". Nikolai Ikhalainen found that this problem happens only in versions 5.5.x and 5.6.x, so chances to see it fixed are low. But I still want to know if this ever happens.
  • Bug #89696 - "Cyclic dependencies are not resolved properly with cascade removal". Make sure to check nice discussion that my dear friend Sinisa Milivojevic had with a bug reporter, Andrei Anishchenko, before marking the bug as "Verified". This regression was most likely caused by a change in MySQL 5.7.21:
    "InnoDB: An iterative approach to processing foreign cascade operations resulted in excessive memory use. (Bug #26191879, Bug #86573)"
  • Bug #89625 - "please package the debug symbols *.pdb files!". Shane Bester always cared about having a way to debug on Windows. Recently I also started to care about this...
It's April Fools' Day today, so why not to make fool of myself assuming that anyone cares about the series of blog posts.

1 comment:

  1. Would it be funny to close the XA bug with the reason XA != HA