This week in Support was busy enough for me. Among other things I had to study all possible reasons (other than obvious query cache impact) for queries hanging in "query end" status and noted Bug #80652 related to binlog group commit and fixed in MySQL 5.7.17+ and 8.0.1+ only. The case I had to review was related to Galera though, and I suggest you to note that "query end" may be related to Galera replication stall. Studying this path further soon brought lp:1197771 - "Cluster stalls while distributing transaction" to my attention again, so I asked about proper status for it on Facebook. As it happens way too often recently, I've got few 'Likes" but no further comments, neither from Percona nor from Codership. It seems good old times of my efficient bugs escalations via Facebook had gone...
Does it mean I am going to stop writing and sharing posts about bugs on Facebook? Definitely it does NOT! One of the recent reasons is this post about Bug #86902 that, after being shared by me, got a quick comment and clarification by Sveta Smirnova that the problem (performance problem when binary logging is turned OFF and you use more than one transactional storage engine concurrently) is actually know, - it was reported as Bug #80870 (that is still "Open", what a shame).
So, even if some Oracle or Percona employees will continue to ignore new bugs reports or my Facebook (re-)posts about MySQL bugs, at least bug reporter gets more chances to get help from Community, additional references and clarifications. So, stay tuned - I'll keep writing about MySQL bugs in this blog and, on a more regular basis and probably for a wider audience, - on Facebook. As popular Mahatma Gandhi's quote says:
I had reported one bug myself this week, Bug #86930 - "Wrong results for queries with row constructors and information_schema". It was quickly verified by Miguel Solorzano. It's a funny bug (with workarounds), but not as scary as few other bugs recently reported and described in this post by Jean-François Gagné.Note, that as he stated, there are more than just that 2 bugs he mentioned (Bug #86926 and Bug #86927). Everybody in Oracle who should know that already knows it, and I know it. I keep stating that the implementation of InnoDB persistent statistics gives us more troubles than benefits. Last time I cared and complained that much it was about Bug #70629 - "InnoDB updated rec_per_key[] statistics not published to the optimizer enough often". But, trust me, this recent set of problems Jean-François found is way more serious.
I'd like to share few references I found extremely useful this week while working on Support issues this week. First of all, if you set up SSL connections for MySQL, take time to review this old post by my former colleague from Percona Roman Vynar, and another one by Matthew Boehm. This may save you some time and debugging efforts. If you care about MariaDB ColumnStore performance (I have to), check this KB article. If you want to replicate data to the table with somewhat different column types on slave, make sure to read the manual. Finally, if you care enough to keep SELinux enabled and use Galera clusters with any non-default ports or directories, make sure to read this blog post.
I've noted a couple of serious decisions made by Oracle in regard to MySQL. First of the is to discontinue work on MySQL Internals manual and integrate it some how into the main manual. You can find related text added recently to some bug reports, like Bug #67989 - "MySQL Internals documentation missing 5.6 binlog protocol parts" by my colleague Andrew Hutchings:
The other one, that is going to be accepted way better, IMHO, is to discontinue work on query cache. Check Bug #86046 - "FROM sub query is cached by mistake" by Meiji Kimura, for example:
Does it mean I am going to stop writing and sharing posts about bugs on Facebook? Definitely it does NOT! One of the recent reasons is this post about Bug #86902 that, after being shared by me, got a quick comment and clarification by Sveta Smirnova that the problem (performance problem when binary logging is turned OFF and you use more than one transactional storage engine concurrently) is actually know, - it was reported as Bug #80870 (that is still "Open", what a shame).
So, even if some Oracle or Percona employees will continue to ignore new bugs reports or my Facebook (re-)posts about MySQL bugs, at least bug reporter gets more chances to get help from Community, additional references and clarifications. So, stay tuned - I'll keep writing about MySQL bugs in this blog and, on a more regular basis and probably for a wider audience, - on Facebook. As popular Mahatma Gandhi's quote says:
"First they ignore you, then they laugh at you, then they fight you, then you win."In my case with public MySQL bugs and the way they are processed by Oracle it was not strictly in this order, and I am yet to see people laughing at me for this reason, but I think I already won as since September 2012 public MySQL bugs database still exists and several Oracle engineers still pay attention to community bug reports there.
I had reported one bug myself this week, Bug #86930 - "Wrong results for queries with row constructors and information_schema". It was quickly verified by Miguel Solorzano. It's a funny bug (with workarounds), but not as scary as few other bugs recently reported and described in this post by Jean-François Gagné.Note, that as he stated, there are more than just that 2 bugs he mentioned (Bug #86926 and Bug #86927). Everybody in Oracle who should know that already knows it, and I know it. I keep stating that the implementation of InnoDB persistent statistics gives us more troubles than benefits. Last time I cared and complained that much it was about Bug #70629 - "InnoDB updated rec_per_key[] statistics not published to the optimizer enough often". But, trust me, this recent set of problems Jean-François found is way more serious.
I'd like to share few references I found extremely useful this week while working on Support issues this week. First of all, if you set up SSL connections for MySQL, take time to review this old post by my former colleague from Percona Roman Vynar, and another one by Matthew Boehm. This may save you some time and debugging efforts. If you care about MariaDB ColumnStore performance (I have to), check this KB article. If you want to replicate data to the table with somewhat different column types on slave, make sure to read the manual. Finally, if you care enough to keep SELinux enabled and use Galera clusters with any non-default ports or directories, make sure to read this blog post.
I've noted a couple of serious decisions made by Oracle in regard to MySQL. First of the is to discontinue work on MySQL Internals manual and integrate it some how into the main manual. You can find related text added recently to some bug reports, like Bug #67989 - "MySQL Internals documentation missing 5.6 binlog protocol parts" by my colleague Andrew Hutchings:
"No more updates are made to the MySQL Internals documentation, because it's in the process of being replaced by https://dev.mysql.com/doc/dev/mysql-server/latest/."Sounds promising, isn't it?
The other one, that is going to be accepted way better, IMHO, is to discontinue work on query cache. Check Bug #86046 - "FROM sub query is cached by mistake" by Meiji Kimura, for example:
"[7 Jul 9:09] Erlend DahlSo, that were some of the links I've noted during this week. Some of the topics mentioned above may be continued and presented in more details in my further posts.
MySQL will no longer invest in the query cache, see: