mysql-proxy-discuss team mailing list archive

[Jan Kneschke <jan@xxxxxxxxxxx>]

Boehm, Matthew wrote:
> Hey Joshua,
> 
>   Thanks. I saw that script and was going to use it originally but I was
> concerned about memory usage of mysql-proxy if it was suddenly asked to
> start maintaining an in-memory record of the analysis of possibly
> hundreds-of-thousands of queries on a daily basis.

This sounds like

  http://mysql.com/products/enterprise/query.html

We use mysql-proxy and script that is based on the analyze-query script
referenced below and dump the created dataset once in a while to our
central repository.

That way we get both:
* reasonable memory usage
* nice statistics across the whole dataset

> I was hoping to turn this into a ‘feature’ we offer to our customers.
> Leave the analysis running all the time and periodically spit out
> reports to our internal customer portal.

Perhaps you just want to integrate with above product.

> MSSQL has some amazing stats that are “always on” and I’m trying to
> implement most of them.
> 
>  
> 
> Another idea I just had; leave the slow_log as it is and basically
> ‘dump’ that table through m-p and let it tokenize into another table.
> That possible? Or would that just be a normal LUA script? (ie: is the
> tokenizer part of m-p or is it pure LUA?)

The tokenizer is a c-library that is wrapper into a lua-module. So you
can use it directly on C-level or in lua, whatever you prefer.

> Hm. Another idea. Continue to use slow_log. Start m-p with a simple
> script that looks for a custom command (‘START SLOW ANALYSIS’, etc)
> which then does a select from slow log, and uses read_result to tokenize
> each row and create the report. May take a while depending on size of
> slow_log but then that could be scripted in a cron job or something.

"mysqldumpslow" does that already, just in pure-perl.

> -Matthew

cheers,
  Jan
-- 
 jan: "Gee, Brain^WEric, what'd you wanna do tonight?"
eric: Same thing we do everynight: Take over the HelloWorld!