Sunday, February 11, 2018

Fun with Bugs #61 - On MySQL Bug Reports I am Subscribed to, Part III

Since my previous post on this topic I've subscribed to 19 more MySQL bugs, so it's time for yet another review of these reports. I am trying to pick up important, funny or hard to process reports every day, and here is the list of the most interesting ones starting from the latest:
  • Bug #89607 - "MySQL crash in debug, PFS thread not handling singals." We have a patch contributed by Robert Golebiowski.
  • Bug #89583 - "no rpm build instructions from source git tree". As Simon Mudd put it:
    "... The sources are supposed to be in the git tree so just tell me how I can use that to produce the files needed to make the rpm SOURCES.

    Without that information only Oracle can build src rpms and no-one else can repeat the process they use and any changes required by people building rpms can't be done in a separate branch of a cloned copy of github.com/mysql/mysql-server. That would be better as from there I can much more easily provide PRs which should improve the MySQL sources."
    I hope one day Oracle will start to share all the code for open source MySQL, properly, including all test cases and detailed, repeatable build instructions. Having proper, up to date and working build instructions that let one end up with binaries built the same way as vendor builds them is not an easy task in general, and KB articles like this is just a step towards the goal...
  • Bug #89559 - "P_S recording wrong digest/digest_text for select statements using views". Everybody knows how much I like Performance Schema, but I like to notice bugs in it even more This is a funny bug that is still "Open" (even though it was noticed by my dear friend Sinisa Milivojevic already) for some reason.
  • Bug #89534 - "Crash during innodb recovery when working with encryption". Yet another bug report (this time "Verified", in 5.7.21) from Robert Golebiowski. Purge thread running during startup may cause problems.
  • Bug #89519 - "Documentation for SSL/TLS and replication is incomplete for 8.0". In this report  DaniĆ«l van Eeden noted that --ssl-verify-server-cert option is not documented.
  • Bug #89444 - "8.0.4rc --initialize-insecure is prohibitively slow". I've noted the same immediately while working on tests for my planned presentation at FOSDEM, but I do not care much about MySQL 8. Fortunately,  Roel Van de Paar cares, so we have a "Verified" bug report about this new behavior.
  • Bug #89441 - "Foreign keys constraints ignored after RENAME TABLE". I am sure we'll see many regressions related to the new data dictionary in MySQL 8. This bug report by Carlos Salguero highlights one of them.
  • Bug #89430 - "Release notes are missing important CVE fixes". Oracle and transparency in anything related to security issues seems to be historically incompatible things. Still, Roel Van de Paar has a hope they may improve their release notes... In the meantime check this great document with all the relevant details.
  • Bug #89375 - "Parallel replication always fails with specific workload from sysbench". It's always great to see public bug report from Oracle employee. Thank you, Frederic Descamps, for sharing your findings with community!
  • Bug #89372 - "Using --no-dd-upgrade seems to have no effect". This bug report by Geert Vanderkelen is still "Open" for some reason.
  • Bug #89367 - "Storing result in a variable(UDV) causes query on a view to use derived tables". Nice optimizer bug reported by Jaime Sicam.
  • Bug #89331 - "Changing lock/release order in group commit causes a deadlock". I think there are at most 10 people in the world who understands all the details of this bug report by Aliaksei Sandryhaila (and I am not one of them). The goal they are trying to achieve is interesting:
    "We are implementing START TRANSACTION WITH CONSTISTENT INNODB SNAPSHOT functionality in 8.0, and want to change it to first taking the next stage's lock, then releasing the previous stage's lock. The implementation of this feature for 5.6 is here: https://github.com/facebook/mysql-5.6/commit/c2b8ced ..."
    I see active discussion with Oracle developer in that report, so I am sure they'll figure out how to proceed.
  • Bug #89272 - "Binlog and Engine become inconsistent when binlog cache file gets out of space". Great finding by Yoshinori Matsunobu and Jeff Jiang. If you ever had "Errcode: 28 - No space left on device" on master while doing some large transactions, make sure to check if slave is in sync. It may be NOT.
  • Bug #89267 - "Unable to access 8.0.4 server after starting on top of 5.7.20 database". I am sure we'll get a lot more bug reports when users actively start to upgrade. For now check comments for possible workaround.
  • Bug #89247 - "Deadlock with MTS when slave_preserve_commit_order = ON." There are many problems with parallel replication/multi-threaded slaves, and Jean-François Gagné probably knows more about them than anyone else. Check also his another bug report about this feature, Bug #89141 - "Error in Group Replication caused by bad Write Set tracking."
  • Bug #89126 - "create table panic on innobase_parse_hint_from_comment". Nice bug report and patch from Yan Huang.
  • Bug #89101 - "MySQL server hang when gtid_mode=on and innodb_thread_concurrency>0". This is a second reason for me to think twice from now on before suggesting to limit innodb_thread_concurrency ever again. Great bug report by Seunguck Lee and MTR test case (and argumentation) by Przemyslaw Malkowski! This is a masterpiece of bug reporting art, Sinisa had to give up...
  • Bug #89098 - "Adding an auto_increment column to existing table creates gaps". Really weird behavior noticed by Riccardo Pizzi. I hope it is truly repeatable literally (but I had not checked myself).
That's all new bug reports in my list of subscriptions. Next time I hope to review few older ones that were not presented in this series yet.