mark - pacemaker list | 2 Aug 2012 01:20
Picon

Quick question regarding wiki vs 'Clusters from Scratch v2'

Good afternoon,


Looking at the corosync configuration examples on the wiki ( http://www.clusterlabs.org/wiki/Initial_Configuration ) and in the Clusters from Scratch document ( http://www.clusterlabs.org/doc/en-US/Pacemaker/1.1/html/Clusters_from_Scratch/_sample_corosync_configuration.html ), there are a number of changes or omissions of settings.

The wiki example includes:

       token: 5000
        token_retransmits_before_loss_const: 20
        join: 1000
        consensus: 7500
        vsftype: none
        max_messages: 20
        clear_nod_high_bit: yes

... and a few other settings that are no longer in the CfSv2 document.  The corosync.conf example there is very close to the default configuration when building from source.  It no longer covers creating a service definition in /etc/corosync/services.d/, because that's no longer needed and instead corosync and pacemaker are started individually.  What about the other settings, though?  I noticed with the CfSv2 sample config only (altered for my network), I get negative node ids since it doesn't include the 'clear_node_high_bit' setting.  Does that have no negative repercussions with pacemaker?  Are any of those other more conservative timing settings still relevant, or is it pretty safe to just tweak the example configuration and let corosync and pacemaker do their thing with largely default values?

Thanks,
Mark
_______________________________________________
Pacemaker mailing list: Pacemaker@...
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://bugs.clusterlabs.org
Andrew Beekhof | 6 Aug 2012 06:53
Favicon
Gravatar

Re: Quick question regarding wiki vs 'Clusters from Scratch v2'

On Thu, Aug 2, 2012 at 9:20 AM, mark - pacemaker list
<m+pacemaker@...> wrote:
> Good afternoon,
>
> Looking at the corosync configuration examples on the wiki (
> http://www.clusterlabs.org/wiki/Initial_Configuration ) and in the Clusters
> from Scratch document (
> http://www.clusterlabs.org/doc/en-US/Pacemaker/1.1/html/Clusters_from_Scratch/_sample_corosync_configuration.html
> ), there are a number of changes or omissions of settings.
>
> The wiki example includes:
>
>        token: 5000
>         token_retransmits_before_loss_const: 20
>         join: 1000
>         consensus: 7500
>         vsftype: none
>         max_messages: 20
>         clear_nod_high_bit: yes
>
> ... and a few other settings that are no longer in the CfSv2 document.  The
> corosync.conf example there is very close to the default configuration when
> building from source.  It no longer covers creating a service definition in
> /etc/corosync/services.d/, because that's no longer needed and instead
> corosync and pacemaker are started individually.  What about the other
> settings, though?  I noticed with the CfSv2 sample config only (altered for
> my network), I get negative node ids since it doesn't include the
> 'clear_node_high_bit' setting.  Does that have no negative repercussions
> with pacemaker?

Pacemaker uses uint32_t for storing the nodeid, so no it doesn't care.
It does make the logs annoying though since not everyone uses %u
everywhere (which correctly interprets the first bit to NOT indicate a
negative number).
And if you're using OCFS2 and possibly GFS2, its still a good idea.

> Are any of those other more conservative timing settings
> still relevant, or is it pretty safe to just tweak the example configuration
> and let corosync and pacemaker do their thing with largely default values?

Pretty much.
The conservative timings came from a time when I was running clusters
with 8+ nodes.
Most people aren't doing that and its a few less things for admins to
think about :)

_______________________________________________
Pacemaker mailing list: Pacemaker@...
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://bugs.clusterlabs.org


Gmane