[Pacemaker] [Question] About "quorum-policy=freeze" and "promote".

[renayama19661014 at ybb.ne.jp]

Hi Andrew,

Thank you for comment.

> > Is it responsibility of the resource agent side to prevent a state of these plural Master?
> 
> No.
> 
> In this scenario, no nodes have quorum and therefor no additional instances should have been promoted.  Thats the definition of "freeze" :)
> Even if one partition DID have quorum, no instances should have been promoted without fencing occurring first.

Okay.
I wish this problem is revised by the next release.

Many Thanks!
Hideo Yamauchi.

--- On Fri, 2014/5/9, Andrew Beekhof <andrew at beekhof.net> wrote:

> 
> On 8 May 2014, at 1:37 pm, renayama19661014 at ybb.ne.jp wrote:
> 
> > Hi All,
> > 
> > I composed Master/Slave resource of three nodes that set quorum-policy="freeze".
> > (I use Stateful in Master/Slave resource.)
> > 
> > ---------------------------------
> > Current DC: srv01 (3232238280) - partition with quorum
> > Version: 1.1.11-830af67
> > 3 Nodes configured
> > 9 Resources configured
> > 
> > 
> > Online: [ srv01 srv02 srv03 ]
> > 
> > Resource Group: grpStonith1
> >     prmStonith1-1      (stonith:external/ssh): Started srv02 
> > Resource Group: grpStonith2
> >     prmStonith2-1      (stonith:external/ssh): Started srv01 
> > Resource Group: grpStonith3
> >     prmStonith3-1      (stonith:external/ssh): Started srv01 
> > Master/Slave Set: msPostgresql [pgsql]
> >     Masters: [ srv01 ]
> >     Slaves: [ srv02 srv03 ]
> > Clone Set: clnPingd [prmPingd]
> >     Started: [ srv01 srv02 srv03 ]
> > ---------------------------------
> > 
> > 
> > Master resource starts in all nodes when I interrupt the internal communication of all nodes.
> > 
> > ---------------------------------
> > Node srv02 (3232238290): UNCLEAN (offline)
> > Node srv03 (3232238300): UNCLEAN (offline)
> > Online: [ srv01 ]
> > 
> > Resource Group: grpStonith1
> >     prmStonith1-1      (stonith:external/ssh): Started srv02 
> > Resource Group: grpStonith2
> >     prmStonith2-1      (stonith:external/ssh): Started srv01 
> > Resource Group: grpStonith3
> >     prmStonith3-1      (stonith:external/ssh): Started srv01 
> > Master/Slave Set: msPostgresql [pgsql]
> >     Masters: [ srv01 ]
> >     Slaves: [ srv02 srv03 ]
> > Clone Set: clnPingd [prmPingd]
> >     Started: [ srv01 srv02 srv03 ]
> > (snip)
> > Node srv01 (3232238280): UNCLEAN (offline)
> > Node srv03 (3232238300): UNCLEAN (offline)
> > Online: [ srv02 ]
> > 
> > Resource Group: grpStonith1
> >     prmStonith1-1      (stonith:external/ssh): Started srv02 
> > Resource Group: grpStonith2
> >     prmStonith2-1      (stonith:external/ssh): Started srv01 
> > Resource Group: grpStonith3
> >     prmStonith3-1      (stonith:external/ssh): Started srv01 
> > Master/Slave Set: msPostgresql [pgsql]
> >     Masters: [ srv01 srv02 ]
> >     Slaves: [ srv03 ]
> > Clone Set: clnPingd [prmPingd]
> >     Started: [ srv01 srv02 srv03 ]
> > (snip)
> > Node srv01 (3232238280): UNCLEAN (offline)
> > Node srv02 (3232238290): UNCLEAN (offline)
> > Online: [ srv03 ]
> > 
> > Resource Group: grpStonith1
> >     prmStonith1-1      (stonith:external/ssh): Started srv02 
> > Resource Group: grpStonith2
> >     prmStonith2-1      (stonith:external/ssh): Started srv01 
> > Resource Group: grpStonith3
> >     prmStonith3-1      (stonith:external/ssh): Started srv01 
> > Master/Slave Set: msPostgresql [pgsql]
> >     Masters: [ srv01 srv03 ]
> >     Slaves: [ srv02 ]
> > Clone Set: clnPingd [prmPingd]
> >     Started: [ srv01 srv02 srv03 ]
> > ---------------------------------
> > 
> > I think even if the cluster loses Quorum, being "promote" the Master / Slave resource that's specification of Pacemaker.
> > 
> > Is it responsibility of the resource agent side to prevent a state of these plural Master?
> 
> No.
> 
> In this scenario, no nodes have quorum and therefor no additional instances should have been promoted.  Thats the definition of "freeze" :)
> Even if one partition DID have quorum, no instances should have been promoted without fencing occurring first.
> 
> > * I think that drbd-RA has those functions.
> > * But, there is no function in Stateful-RA.
> > * As an example, I think that the mechanism such as drbd is necessary by all means when I make a resource of Master/Slave newly.
> > 
> > Will my understanding be wrong?
> > 
> > Best Regards,
> > Hideo Yamauchi.
> > 
> > 
> > _______________________________________________
> > 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://bugs.clusterlabs.org
> 
>