Sergey,
I tested the tree as much as I could and I filed all the bugs that do not
appear duplicates. In order to test further, the existing bugs need to be
fixed, in particular bug 665669 which causes random sporadic failures that
may mask other bugs.
If possible, for the next round of testing a single unified tree should be
used that contains all fixes from all other trees (including those from
maria-5.3), and the maria-5.3-mwl128 and maria-5.3-dsmrr-cpk trees should be
retired.
Note that in bug 671489 the EXPLAIN does not mention MRR at all, while the
result set is influenced by MRR options. We need to avoid such situations as
much as possible -- the action of every optimizer_switch should be reflected
in the EXPLAIN.
I am having a situation with InnoDB where a bug is perfectly reproducible in
the RQG, but not when an MTR test case is extracted, even if I use optimizer
hints to force the MTR test case to produce the same query plan. It appears
to me that this is because Innodb row estimates fluctuate constantly
throughout the lifetime of the server. Igor says that InnoDB should not be
required to reproduce any bug, however my experience suggests otherwise. Any
ideas on how to make MTR test cases such bugs would be much appreciated.
Philip Stoev
----- Original Message -----
From: "Sergey Petrunya"
Hi Philip,
The combined tree with DS-MRR improvements and MWL#128 is ready for testing. It is located at lp:~maria-captains/maria/maria-5.3-mwl128-dsmrr-cpk/ .
I don't know how hard that is for you, but it would be useful to check the failures - against 5.3-main tree (to see whether the problem is in the new code or not) - against lp:~maria-captains/maria-5.3-mwl128 (to see whether the problem is in MWL#128 or in the DS-MRR code)
BR Sergey -- Sergey Petrunia, Software Developer Monty Program AB, http://askmonty.org Blog: http://s.petrunia.net/blog