Comments on: Google’s user_statistics V2 port and changes

By: Baron Schwartz

Baron Schwartz — Tue, 03 Mar 2009 03:22:49 +0000

Get one of our builds. I assume you are using a standard MySQL server. Download our version from http://www.percona.com/percona-lab.html and then see the docs here: http://www.percona.com/docs/wiki/patches:userstatv2

By: True

True — Mon, 02 Mar 2009 20:56:00 +0000

How to apply this patch ? I am using mysql 5.0.77.

I dont see table index_statistics

By: peter

peter — Mon, 15 Sep 2008 16:40:40 +0000

Thanks Kevin

This is exactly the point.

By: Kevin Burton

Kevin Burton — Sat, 13 Sep 2008 18:29:26 +0000

Peter,

I realize it wasn’t the complete patch…. I was congratulating you on isolating one of the features from the larger V2 so that it can be easily added to MySQL AB’s sources and moved between versions easier.

Kevin

By: Mark Callaghan

Mark Callaghan — Sat, 13 Sep 2008 14:50:09 +0000

@Peter — What is CPU time per query? Select_commands counts all SELECT statements. Update_commands counts all INSERT/UPDATE/REPLACE/DELETE statements and Other_commands counts everything else. Cpu_time / (Select_commands + Update_commands + Other_commands) displays time per client-server protocol command. Do you want Cpu_time for each type of command? The expressions are easy to evaluate since you provide this data via the Information Schema.

By: peter

peter — Sat, 13 Sep 2008 05:46:34 +0000

Couple of comments

1) It is interesting how much speed overhead if any we got with these patches.
2) empty_queries is misleading It really means empty_results though I guess it is best if Google renames it so we do not add more differences than needed
3) Busy Time, CPU time all counted with better accuracy while displayed as seconds only this may not be accurate enough in some cases.
4) The patch should be counting CPU time per query which can be very nice addition to the processlist schema besides time_ms we have already.

By: peter

peter — Sat, 13 Sep 2008 05:37:28 +0000

Kevin,

What is showed here is not the Googles V2 complete patch set but the given accounting patch ported to Percona patch set. The user statistics patches in particular are very safe because they do not change any internal logic this is why we put them in the “safe to use” release.

By: Kevin Burton

Kevin Burton — Sat, 13 Sep 2008 04:11:20 +0000

Nice….

I think the biggest obstacle to V2 adoption is to get the patches into smaller bite size chunks.

It’s looking like 5.0.67 is the sweet spot though…..with with the Google SMP patches.

If you can get a build with these patches and your InnoDB performance patches I think this is the release we’ll target.

There are still some cool V2 features though… I’m going to try to blog about them when I get some time.

By: Mark Callaghan

Mark Callaghan — Fri, 12 Sep 2008 22:20:28 +0000

Wow. Who knew patches could be incorporated so quickly. Great work.