maria-developers team mailing list archive

[Sergey Petrunya <psergey@xxxxxxxxxxxx>]

Hi! 

As discussed on the yesterday's call, we're starting to prepare for the
next 5.3 release, 5.3.2 beta. The intent is to build it on this or on the 
next weekend.

Before the release, we need to do
1. Bugs 
2. optimizer_switch changes.
3. Merge from mariadb-5.1

We don't need to address:
1. Pushing Kristian's replication features. If I heard correctly, the 
   decision was that they won't be pushed into 5.3 (please correct me 
   if I am wrong)

== Bugs ==

Current situation with 5.3-targeted bugs is as follows:

Crashing bugs:
- SergeyP: 3 + 2 non-repeatable (#860535 #860553 #861147  
                                 #860561 (non-repeatable) 
                                 #860580 (non-repeatable))
- Timour:  2  (#824425 #858148)
- Nobody:  1  (#859375)

Wrong query result bugs:
- Timour: 8  (#858038 #817966 #833777 #825051 #826150 #747278 #833702 #856152)
- Igor :  2  (#823301 #791761)
- Sanja:  1  (#825075)


I have a slight excuse that all my crasing bugs were filed just yesterday.
I'll try to fix them before the release.

Another problem area is Timour's crashes + wrong query result bugs.

Also, we need to assign #859375.

== optimizer_switch changes ==

Some of the new optimizations are already sufficiently stable so that they can
be made enabled by default.

>From the code quality point of view, Philip's opinion is:
- subquery_cache can be turned ON.
- join_cache_level can be set to value higher than 1, and also 
  various variants of BKA/Hash join may be enabled.
- MRR can be enabled
- index_condition_pushdown can be enabled

- Materialization still has bugs, cannot be enabled.
- semi-joins+materialization cannot be enabled together, there are bugs.
- semi-joins without materialization cause non-repeatable crashes, so 
  cannot be enabled yet.
- FROM subqueries/derived tables optimization didn't recieve sufficient testing
  yet, so cannot be enabled.

Note that the above is based only on the number of known crashes/wrong query
result problems, performance issues are not counted.


My opinion is that
- subquery_cache can be enabled.
- index_condition_pushdown can be enabled.
- I'm hesitant to enable MRR because it may cause slowdowns for small-dataset
  sysbench-like tests.
- perhaps, some part of hash join could be enabled (as far as I understand, its
  overhead is rather small, so it doesn't cause regressions?)

Anybody (and especially feature "owners") have any thoughts?

== Merge from MariaDB 5.1 ==
Monty mentioned we'll need to do it, but nobody took this task?

BR
 Sergey
-- 
Sergey Petrunia, Software Developer
Monty Program AB, http://askmonty.org
Blog: http://s.petrunia.net/blog