[pgpool-general: 7456] Expected behaviour after pcp_detach_node ?

[Anssi Kanninen]

Hi!

What is the expected behaviour when detaching a node?
I have a three-node test cluster with pgpool-II 4.1.3
and PostgreSQL 12 like below.


id | hostname | status |  role   | repl_delay | repl_state | repl_sync_st | last_status_change
  0 | 10.0.0.1 | up     | primary | 0          |            |              | 2021-03-25 12:18:27
  1 | 10.0.0.2 | up     | standby | 0          | streaming  | async        | 2021-03-25 12:18:56
  2 | 10.0.0.3 | up     | standby | 0          | streaming  | async        | 2021-03-25 12:18:56

*** Then I run "pcp_detach_node 0". The second node was promoted as expected:

id | hostname | status |  role   | repl_delay | repl_state | repl_sync_st | last_status_change
  0 | 10.0.0.1 | down   | standby | 0          |            |              | 2021-03-25 12:25:27
  1 | 10.0.0.2 | up     | primary | 0          |            |              | 2021-03-25 12:25:27
  2 | 10.0.0.3 | down   | standby | 0          |            |              | 2021-03-25 12:25:27

*** But then, the first one is back as primary again?

id | hostname | status |  role   | repl_delay | repl_state | repl_sync_st | last_status_change
  0 | 10.0.0.1 | up     | primary | 0          |            |              | 2021-03-25 12:25:36
  1 | 10.0.0.2 | up     | standby | 0          |            |              | 2021-03-25 12:25:36
  2 | 10.0.0.3 | down   | standby | 0          |            |              | 2021-03-25 12:25:27

*** After that, returning as standby, the second node is primary:

id | hostname | status |  role   | repl_delay | repl_state | repl_sync_st | last_status_change
  0 | 10.0.0.1 | up     | standby | 0          |            |              | 2021-03-25 12:25:45
  1 | 10.0.0.2 | up     | primary | 0          |            |              | 2021-03-25 12:25:45
  2 | 10.0.0.3 | up     | standby | 0          |            |              | 2021-03-25 12:25:45

*** After all, this is the final state.

id | hostname | status |  role   | repl_delay | repl_state | repl_sync_st | last_status_change
  0 | 10.0.0.1 | up     | standby | 33463880   |            |              | 2021-03-25 12:25:45
  1 | 10.0.0.2 | up     | primary | 0          |            |              | 2021-03-25 12:25:45
  2 | 10.0.0.3 | up     | standby | 0          | streaming  | async        | 2021-03-25 12:25:45

So finally, the second node was promoted as expected.

But why is the first node "up" again even if it's not replicating?

See pgpool.conf attached.

Best regards,
   Anssi Kanninen




-- 
anssi at iki.fi
-------------- next part --------------
A non-text attachment was scrubbed...
Name: pgpool.conf
Type: application/octet-stream
Size: 44465 bytes
Desc: 
URL: <http://www.pgpool.net/pipermail/pgpool-general/attachments/20210325/ac32b97e/attachment-0001.obj>