Pgsql resource agent of pacemaker

On Mon, 19 Aug 2019, 18:47 Shital A, <brightuser2019@gmail.com> wrote:

Hello,

Need advise on below situation:

Postgres 9.6
Pacemaker 1.1.19
Corosync 2.4.3

We are testing HA setup on a two node cluster using pacemaker, corosync
stack. The replication is streaming replication in async mode.
Whenever there is a failover to standby, pgsql created a PGSQL.lock file
in /var/lib/pgsql/tmp/ folder. We noticed that this file is created at
random on any one of the nodes.

Want to know:
- Is there any logic behind which node it creates the lock file?

- When a node in kept in HS:alone state ?

- How can we check if failed node is at a log location behind current
primary? So that we can force RA to start by deleting lock file?

Please help to understand

Thanks.

Any thoughts?

Thanks in advance!

On Mon, 19 Aug 2019, 23:36 Shital A, <brightuser2019@gmail.com> wrote:

On Mon, 19 Aug 2019, 18:47 Shital A, <brightuser2019@gmail.com> wrote:

Hello,

Need advise on below situation:

Postgres 9.6
Pacemaker 1.1.19
Corosync 2.4.3

We are testing HA setup on a two node cluster using pacemaker, corosync
stack. The replication is streaming replication in async mode.
Whenever there is a failover to standby, pgsql created a PGSQL.lock file
in /var/lib/pgsql/tmp/ folder. We noticed that this file is created at
random on any one of the nodes.

Want to know:
- Is there any logic behind which node it creates the lock file?

- When a node in kept in HS:alone state ?

- How can we check if failed node is at a log location behind current
primary? So that we can force RA to start by deleting lock file?

Please help to understand

Thanks.

Any thoughts?

Thanks in advance!

Hello,

Was wandering if this is not the right mailing list for this issue. Please
let me know.

Pacemaker on pgsql do not behave in a fixed pattern. The lock file gets
created on any node. According to link
https://wiki.clusterlabs.org/wiki/PgSQL_Replicated_Cluster it should get
created on the failing node however we are seeing it being created on the
promoted node.

If you have such setup in production or have evaluated, please provide
inputs.

Thanks.