On 17/07/2013, at 11:38 AM, Jeff Frost <[email protected]> wrote: > I understand that it can't be demoted in the normal sense, I just expected it > to demote after migration-threshold failures were reached and not always on > the first failure.
Its a bit more subtle than you might imagine... We don't just demote and re-promote a resource when the master fails. It goes the through the full stop/start cycle too - which probably updates the promotion scores - which probably requires the PE to do some recalculations at a point the old master is stopped - at which point the promotion score for the old master is (and must) be ignored - which means a different node gets promoted instead So its a reasonable expectation, its just not one we can currently live up to unfortunately. > > BTW, the RA appears to work great, so thanks for your work in this regard! > > > On Jul 16, 2013, at 5:17 PM, Takatoshi MATSUO <[email protected]> wrote: > >> Hi Jeff >> >> The RA stops PostgreSQL on demote because PostgreSQL can't demote. >> So it makes no sense to set migration-threshold when using pgsql RA >> with replication. >> >> I tried to restart PostgreSQL on demote, but it may cause data inconsistency. >> It needs complicated mechanism to avoid it. >> >> >> They say that PostgreSQL 9.4 may be able to avoid it in the future. >> >> Thanks, >> Takatoshi MATSUO >> >> 2013/7/17 Jeff Frost <[email protected]>: >>> Since I've been playing with the pgsql resource agent, it seems as though >>> migration-threshold is ignored on the master node. >>> >>> That is, when I stop the postgresql service on the master, it is immediately >>> migrated even though migration-threshold is set to "30" in both the pgsql >>> primitive and msPostgresql ms. >>> >>> On the slave, it appears to act as expected. >>> >>> Is it expected to act differently on the master or is there a different meta >>> tag I should be using for master/slave clones? > > > > > > > _______________________________________________ > 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
