[pgpool-general: 8407] Re: failover

[Bo Peng]

Hello,
 
> > In a properly configured pgpool cluster should one be able to failover a 2nd time in a 3 node cluster?  To be clear, systemA is primary and fails, leaving systemB (new primary) and systemC (standby) up and running.  Shortly after, SystemB fails.  Should the VIP move to systemC and continue to allow RW access to the Data?
> 
> No.
> If only SystemB and SystemC are running and SystemB fails,
> VIP will not be assigned to any node, because quorum doesn't exist.
> It is the safest way to avoid split brain.

Above is the default behavior.

If you are using Pgpool-II 4.3, you can enable 

wd_remove_shutdown_nodes
OR
wd_lost_node_removal_timeout

to remove a shutdown/lost node from watchdog cluster.
In this case, VIP will be assigned even if only one Pgpool-II node is running.
Note that using these parameters may cause split brain.

For detailed information, see "5.14.7. Controlling the watchdog cluster membership":
https://www.pgpool.net/docs/latest/en/html/runtime-watchdog-config.html#CONFIG-WATCHDOG-CLUSTER-MEMBERSHIP

-- 
Bo Peng <pengbo at sraoss.co.jp>
SRA OSS LLC
https://www.sraoss.co.jp/