[Pacemaker] Adding VIP support for the MySQL RA

Yves Trudeau y.trudeau at videotron.ca
Sat Nov 12 13:22:25 EST 2011


lol... How many large databases have you managed?  Once evicted, MySQL 
will be restarted by Pacemaker so all the caches will be cold.  On a 
128GB db server with an InnoDB buffer pool of >100 GB, warming the cache 
takes hours and during that time the db is much slower.  Since the slave 
has been evicted because it was lagging, restarting it will not help at all.

Regards,

Yves



On 11-11-12 12:24 PM, Michael Marrotte wrote:
> Why's evicted terrible?  It optional and simply removes the slave node 
> from the cluster if it's lagging too far behind the master.  The 
> collocation IP can float to another slave in the cluster or go 
> offline.  I thought that's what you're looking for.  I can post a 
> config if you're interested.
>
> Sent from my iPhone
>
> On Nov 12, 2011, at 11:51 AM, Yves Trudeau <y.trudeau at videotron.ca 
> <mailto:y.trudeau at videotron.ca>> wrote:
>
>> Hi,
>>    Setting evict_outdated_slaves is terrible but then how do you 
>> remove the read VIPs from slaves that are outdated.  If there is a 
>> way to do it, tell me :)  I am always open to do things in a simpler 
>> way.  This solution works without evicting outdated slaves.   The 
>> current mysql RA does behave correctly HA wise but it is less than 
>> optimal database wise.  I mean in small installations it is probably 
>> ok, but at Percona, I hit many high end installations where the 
>> current behavior is not adequate.  I consider myself more as a 
>> database specialist than an HA one so I may be more sensitive to 
>> these problems.
>>
>> Regards,
>>
>> Yves
>>
>> On 11-11-12 09:19 AM, Michael Marrotte wrote:
>>> Are you using evict_outdated_slaves?
>>> <parameter name="evict_outdated_slaves" unique="0" required="0">
>>>
>>> <longdesc lang="en">
>>> If set to true, any slave which is more than max_slave_lag seconds
>>>
>>> behind the master has its MySQL instance shut down. If this parameter
>>> is set to false in a primitive or clone resource, it is simply
>>> ignored. If set to false in a master/slave resource, then exceeding
>>>
>>> the maximum slave lag will merely push down the master preference so
>>> the lagging slave is never promoted to the new master.
>>> </longdesc>
>>> <shortdesc lang="en">Determines whether to shut down badly lagging
>>> slaves</shortdesc>
>>>
>>> <content type="boolean" 
>>> default="${OCF_RESKEY_evict_outdated_slaves_default}" />
>>> </parameter>
>>>
>>>
>>> On Fri, Nov 11, 2011 at 9:57 PM, Yves Trudeau 
>>> <y.trudeau at videotron.ca <mailto:y.trudeau at videotron.ca>> wrote:
>>>
>>>     Because that's not enough, if a slave lags behind too much, I
>>>     want to remove the vip but not stop the slave.
>>>
>>>     On 11-11-11 06:12 PM, Michael Marrotte wrote:
>>>>     Why don't you simply colocate VIP's with the master/slave roles?
>>>>
>>>>     On Fri, Nov 11, 2011 at 5:18 PM, Yves Trudeau
>>>>     <y.trudeau at videotron.ca <mailto:y.trudeau at videotron.ca>> wrote:
>>>>
>>>>         Hi,
>>>>           I created a fork of the resource-agents and modified the
>>>>         mysql RA to supports VIP.  The support for read/write VIPs
>>>>         is something that currently limits the use of pacemaker to
>>>>         manage mysql replication.  I ran some basic tests on the
>>>>         modified agent and it seems behaving sanely so far but more
>>>>         tests will be needed (next week).  I also modified a few
>>>>         things in the RA that were not optimal in my view.   Here's
>>>>         the URL for my fork:
>>>>
>>>>         git://github.com/y-trudeau/resource-agents.git
>>>>         <http://github.com/y-trudeau/resource-agents.git>
>>>>         https://y-trudeau@github.com/y-trudeau/resource-agents.git
>>>>
>>>>         In order to use it, you need to adapt the pacemaker
>>>>         configuration, here's the one I am using:
>>>>
>>>>         http://pastebin.com/QPJPWDhx
>>>>
>>>>         Next, apart from running more tests,  I want to do the
>>>>         following:
>>>>
>>>>         - review user privileges used, need to lower the ones of
>>>>         replication_user
>>>>         - add support for MHA
>>>>          (http://code.google.com/p/mysql-master-ha/)
>>>>         - run a beautifier for more uniformity in style
>>>>
>>>>         Reviews and comments are of course welcome.
>>>>
>>>>         Regards,
>>>>
>>>>         Yves
>>>>
>>>>
>>>>
>>>>
>>>>         _______________________________________________
>>>>         Pacemaker mailing list: Pacemaker at oss.clusterlabs.org
>>>>         <mailto:Pacemaker at oss.clusterlabs.org>
>>>>         http://oss.clusterlabs.org/mailman/listinfo/pacemaker
>>>>
>>>>         Project Home: http://www.clusterlabs.org
>>>>         Getting started:
>>>>         http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf
>>>>         Bugs:
>>>>         http://developerbugs.linux-foundation.org/enter_bug.cgi?product=Pacemaker
>>>>
>>>>
>>>>
>>>>     _______________________________________________
>>>>     Pacemaker mailing list:Pacemaker at oss.clusterlabs.org  <mailto:Pacemaker at oss.clusterlabs.org>
>>>>     http://oss.clusterlabs.org/mailman/listinfo/pacemaker
>>>>
>>>>     Project Home:http://www.clusterlabs.org
>>>>     Getting started:http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf
>>>>     Bugs:http://developerbugs.linux-foundation.org/enter_bug.cgi?product=Pacemaker
>>>
>>>
>>>     _______________________________________________
>>>     Pacemaker mailing list: Pacemaker at oss.clusterlabs.org
>>>     <mailto:Pacemaker at oss.clusterlabs.org>
>>>     http://oss.clusterlabs.org/mailman/listinfo/pacemaker
>>>
>>>     Project Home: http://www.clusterlabs.org
>>>     Getting started:
>>>     http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf
>>>     Bugs:
>>>     http://developerbugs.linux-foundation.org/enter_bug.cgi?product=Pacemaker
>>>
>>>
>>>
>>> _______________________________________________
>>> Pacemaker mailing list:Pacemaker at oss.clusterlabs.org
>>> http://oss.clusterlabs.org/mailman/listinfo/pacemaker
>>>
>>> Project Home:http://www.clusterlabs.org
>>> Getting started:http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf
>>> Bugs:http://developerbugs.linux-foundation.org/enter_bug.cgi?product=Pacemaker
>>
>> _______________________________________________
>> Pacemaker mailing list: Pacemaker at oss.clusterlabs.org 
>> <mailto:Pacemaker at oss.clusterlabs.org>
>> http://oss.clusterlabs.org/mailman/listinfo/pacemaker
>>
>> Project Home: http://www.clusterlabs.org
>> Getting started: http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf
>> Bugs: 
>> http://developerbugs.linux-foundation.org/enter_bug.cgi?product=Pacemaker
>
>
> _______________________________________________
> Pacemaker mailing list: Pacemaker at oss.clusterlabs.org
> http://oss.clusterlabs.org/mailman/listinfo/pacemaker
>
> Project Home: http://www.clusterlabs.org
> Getting started: http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf
> Bugs: http://developerbugs.linux-foundation.org/enter_bug.cgi?product=Pacemaker


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.clusterlabs.org/pipermail/pacemaker/attachments/20111112/5414f766/attachment-0003.html>


More information about the Pacemaker mailing list