The Future of Performance: Scaling

[David Lawrence]

Max Kanat-Alexander wrote:
> On Tue, 30 Dec 2008 22:16:20 -0600 "Bradley Baetz" <bbaetz at acm.org>
> wrote:
>   
>> In general, we scale read-only pretty well. 
>>     
>
> 	Yeah, although there are a few operations that don't take as
> much advantage of that as they could. There are probably a lot more
> places where we could be using the shadow DB.
>
>   
>> The shadowdb stuff pushes db reads off, but Bugzilla doesn't have a
>> clear goal as to how out of date data can be - if we relaxed the rules
>> a bit then some more queries could possibly move to the replication
>> db. However, I don't think that thats much of an issue given the
>> typical read:write ratio.
>>     
>
> 	Yeah, on most installations I don't think it's an issue, but at
> places like bmo it could always be handy to see what else we can push
> off. Apparently the ratio of master reads to slave reads increased a
> lot in 3.2 for some reason, and we should probably look into it.
>
> 	-Max
>   

Now that we allow search engines to index bugzilla.redhat.com using 
sitemap index files, we have a large amount
of accesses to show_bug.cgi over the past month. We have seen, using 
mysql's process list commands, that
a large amount of "DELETE FROM logincookies WHERE TO_DAYS(NOW()) - 
TO_DAYS(lastused) > 30"
statements are being executed that was causing the replicated slaves to 
get backed up. This statement is executed
everytime Bugzilla->login is called without auth data and since DELETE 
is a SELECT/INSERT block for a table, things could get slower if it is 
executed very frequently.

To solve the issue we are moving that code to a separate script that 
will be ran once a night to clean up expired
cookies in the logincookies table. By commenting out that line in 
Bugzilla::Auth we saw the database load decrease and the slaves were 
able to get caught up. Does this sound like something that would be 
useful to the
upstream and a bug/patch generated?

Dave

-- 
David Lawrence, RHCE  dkl at redhat.com
------------------------------------
Red Hat, Inc.    Web: www.redhat.com
1801 Varsity Drive Raleigh, NC 27606