Comments on: Getting annoyed with MyISAM multiple key caches.

By: Nate Pierce

Nate Pierce — Thu, 21 May 2009 12:07:49 +0000

After seeing the wafflegrid presentation at the conference this year, I am kicking around the idea of trying to replace and/or augment the myisam keycache with memcached. Due to resources, we’re still running on 4.0, so this would get us past the single 4G global keycache barrier w/out having to wade through upgrade issues and buy us some time. I’m not intimately familiar with either codebase, but it looks like it may be possible. Anyone have any thoughts on the matter?

By: peter

peter — Thu, 07 May 2009 17:41:49 +0000

Steven,

Right. Though because you have to map whole indexes it is not so easy – the slave thread vs reads would often touch same tables just different pages.

I find in most useful to deal with MyISAM contention.

By: Steven Roussey

Steven Roussey — Thu, 07 May 2009 15:32:10 +0000

Multiple key caches are great when you know what your problem is and how extra caches can be utilized. Also handy for a server that is also a slave of another, since you can limit the cache dilution of the slave process.

By: Shane Bester

Shane Bester — Tue, 05 May 2009 07:19:59 +0000

should never use ‘mysqladmin debug’ on a server that is even moderately loaded. it tends to crash very often… there are a few bug reports about it already.

By: peter

peter — Sat, 02 May 2009 16:45:31 +0000

Matthew,

Yes this can be implemented as a plugin. I however does not like this plugin obsession. I would like to see the server have features which I need in its default installation (though I do not really care if this is linked in stuff or plugins) – just having plugin available somewhere is helpful for your own installation but if you’re dealing with customer systems it is a lot of hassle – you well need feature when you do not want to restart the server on the first place. Second adding third party plugins may require internal approval and QA process. MyISAM is compiled in so I think such little things should be too.

By: peter

peter — Sat, 02 May 2009 16:41:11 +0000

Harrison,

Thanks – indeed this way I can see the key caches in the error log, though it does not make it easily accessible from the applications. You’re right – locks held is another issue where MyISAM is lacking.

By: Harrison Fisk

Harrison Fisk — Sat, 02 May 2009 15:34:52 +0000

You can see most of this via mysqladmin debug in the error log (except the mappings), so you can know the sizes and tuning and such. I think it would be nice to have an I_S table or similar, but that is true for several things with MyISAM, such as locks.

For example:

Key caches:
default
Buffer_size: 314572800
Block_size: 1024
Division_limit: 100
Age_limit: 300
blocks used: 3
not flushed: 0
w_requests: 0
writes: 0
r_requests: 6
reads: 3

keycache1
Buffer_size: 131072
Block_size: 1024
Division_limit: 100
Age_limit: 300
blocks used: 0
not flushed: 0
w_requests: 0
writes: 0
r_requests: 0
reads: 0

By: Matthew Montgomery

Matthew Montgomery — Sat, 02 May 2009 15:22:34 +0000

Sounds like a perfect job for an information_schema plugin.

http://dev.mysql.com/tech-resources/articles/mysql_i_s_plugins_part1-2.html
http://dev.mysql.com/tech-resources/articles/mysql_i_s_plugins_part2.html

Don’t get annoyed, get busy

By: http://jeremy.zawodny.com/blog/

http://jeremy.zawodny.com/blog/ — Sat, 02 May 2009 14:52:51 +0000

Agreed 100%. It feels like the multiple key cache support was added for someone who desperately needed it and didn’t care about the administrative side at all.

By: Mark Callaghan

Mark Callaghan — Sat, 02 May 2009 14:07:34 +0000

Is Maria still expected to replace MyISAM? It was at one point and that was justification for not fixing performance problems in MyISAM (problems including mutex contention).