Friday, July 26, 2013

Fun with Bugs #19 - waiting for MySQL 5.6.13 and some real fun?

I feel like MySQL 5.6.12 was released ages ago, while in reality it was on June 3, less than 2 months ago. No wonder I feel so, after writing several posts about bugs fixed and not fixed in it... Anyway, we still have to wait for MySQL 5.6.13 for a week or even two probably and in the meantime I decided to write new post for this series based on good old idea of making a digest of my recent bugs-related posts at Facebook. I know, it's boring and annoying (same as waiting for the release of 5.6.13).

Let's start with Bug #69846 - "ICP does not work on UNIQUE indexes". Based on my quick tests presented there I'd say that ICP (index condition pushdown) actually does not work for InnoDB clustered key (being it explicit PRIMARY KEY or just the first UNIQUE KEY), and this is probably by design, as such index is "equal" to InnoDB table itself. I'd still prefer to see this explained in the manual. Bug is still "Open", so Oracle engineers may have different opinion.

Another bug made my day actually, it's Bug #69842 from Domas. It does not matter that it is "Open" or that doublewrite buffer does not work efficiently with InnoDB page sizes smaller than 16K... Question is how to create proper Latin-based term for writing something 3, 5 ("quintuple-writing") or 17 times...

Optimizer in MySQL had always been a source of endless fun. It seems that for some cases it worked properly last time in 5.1 (if not 5.0). Check Bug #69833 for a recent example, "Bad interaction between MIN/MAX and "HAVING SUM(DISTINCT)": wrong results". Yes, it's a weird corner case with not much real life use, but still it's a regression bug. In some areas MySQL 5.1 was better than any 5.5 or 5.6 GA version from Oracle so far :)

Bug #62578 is still affecting customers who use MySQL 5.5.x. Recent checks show that 5.6.12 is NOT affected (neither Sinisa nor me were able to repeat it on 5.6.12), but mysql client of 5.5.32 still crashes when you reside terminal (or putty) window it runs in. Fix is well know and probably found its way to 5.6 long time ago, but why it is NOT in 5.5? Yes, I still remember talks about a policy to fix bugs in older GA versions only when customers ask and there is no risk or high efforts involved, but isn't this the case for this bug?

I've spent a lot of time trying to find similar known bug report for Bug #69825 reported by my colleague, but failed. It seems new problem, at least in public bugs database. Sounds simple, "InnoDB: Assertion failure in thread ... in file line 297", affects 5.6.12 and the only repeatable test case so far is based on customer confidential data (as far as I know), but still... Why it is still "Open" and got no comments? Had everybody in Oracle decided that GTID-based crash safe replication with information stored in InnoDB tables in mysql database just always work, so it must be some bug reporter fault? It's not the case, dear colleagues, it's not the case...

Now something simple and easy to verify by code review, Bug #69827. "Hardcoded libdir in cmake", so that lib64 on Linux is used only for x86_64. Who cares I've asked? But even if we do not take 64-bit PowerPC into account any more, some sources claim 64-bit ARM-based servers are coming soon... Should be easy fix actually (hacked in by many users already probably).

If you are going to upgrade to MySQL 5.6 and had forgotten that configuration file in basedir (if present) overrides settings in /etc/my.cnf, check Bug #68643 - "sql_mode is unkind in my.cnf created by mysql_install_db". Yes, as it silently includes STRICT_TRANS_TABLES. It's just a feature request for now, so hardly is going to change any time soon (if at all). Just take care about this, I warned you.

 Week started with funny bugs mostly. This one is open for more than 7 years and still scare people, Bug #18256. If session is killed you still may see messages like this in your mysql command line client:

ERROR 1053 (08S01) at line 1: Server shutdown in progress

Do not panic, please. Server may be perfectly alive, just your session no longer works.

On the weekend I still had fun with the manual and friends who are scared by Oracle requests for personal information. Check Bug #69805 I've reported (or for the whole story) and note that fine manual does not describe what was really implemented correctly. Let them now decide who was wrong, manual (I hope) or implementation of the feature as designed. Error handling in stored routines is still to complex and far from perfect even in MySQL 5.6 if you ask me, at least comparing to good old Oracle PL/SQL (this is what I really miss since 2005 sometimes...)

That's all for now. I've spent this week digging around and trying to fix things and protect strangers from dangers like this:

Hope this helps somebody.

No comments:

Post a Comment