1. Well, I was re-added slave nodes to cluster, so It work well (now). I will leave cluster idle on all night for repeat this error and show you unclean state and logs
2. Thanls a lot for this advice! But I don't understand one thing - how to switch slave from sync to async mode? Yes, I want all nodes in async. The data loss is not important. 2013/12/11 Takehiro Matsushima <[email protected]> > Sorry for my late reply... > > > 1. Why slaves stay disconnected? How I can found reason? > > It seems that slaves are connected properly. > master-pgsql:1 on Node b has score 100, means it can promote when > current master failed. > > > Node Attributes: > > * Node b.mydomain.com: > > + master-pgsql:0 : -INFINITY > > + master-pgsql:1 : 100 > > + pgsql-data-status : STREAMING|SYNC > > + pgsql-status : HS:sync > > + pgsql-xlog-loc : 000000000E0000C8 > > You can confirm any changes replicate to slaves by; > 1. INSERT on Master > 2. SELECT on one of slave > 3. Same on another one > > > > 2. Why one node stay as sync replication, but other is async? > > I've also been asked to Mr. MATSUO about this thing. > > 1. PostgreSQL has 4 states, Master, Sync, Potential and Async. > 2. Sync slave node is only one, and others are Async. > 3. One of Async slave is able to change Sync when Sync slave failed, > it called Potential. > 4. Changing Potential to Sync is controlled by PostgreSQL, NOT Pacemaker. > > He showed to me the situation following; > > 1. Node A is Master, Node B is Sync slave, Node C is Potential Async slave. > 2. Node B's replication line has down, then PostgreSQL promotes Node C > to Sync slave and responses COMMIT OK to client. > 3. But, Pacemaker doesn't know this event yet. > 4. After that Node A crashed, then Pacemaker tries to promote Node B, > it known Sync slave. > 5. Consistency is broken. > > So, the 3rd node is set Async intentionally, Mr. MATSUO answered. > And also refer followings he said. > > https://github.com/ClusterLabs/resource-agents/commit/55494b5052f540030938733ec4729cc37ac64a8c > https://github.com/t-matsuo/resource-agents/issues/24 > > > -- > Regards, > Takehiro Matsushima > _______________________________________________ > Linux-HA mailing list > [email protected] > http://lists.linux-ha.org/mailman/listinfo/linux-ha > See also: http://linux-ha.org/ReportingProblems > _______________________________________________ Linux-HA mailing list [email protected] http://lists.linux-ha.org/mailman/listinfo/linux-ha See also: http://linux-ha.org/ReportingProblems
