[pgpool-hackers: 4182] Re: failover_on_backend_shutdown not working

[Tatsuo Ishii]

> I am upgrading my PgPool setup from version 4.2.2 to 4.3.2.
> 
> I was testing the new parameter that this version introduces - failover_on_backend_shutdown, but it doesn’t seem to work.
> 
> My setup is 1 PgPool (4.3.2), 1 primary postgres, 2 replica postgres (all 12.5).
> 
> Initially, failover_on_backend_shutdown was turned on, then I turned it off from the config file and did a reload of PgPool.
> 
> The option is shown correctly when I query it in runtime:
> 
> postgres=# pgpool show failover_on_backend_shutdown;
>  failover_on_backend_shutdown
> ------------------------------
>  off
> (1 row)
> 
> postgres=# show pool_version;
>  pool_version
> -----------------------
>  4.3.2 (tamahomeboshi)
> (1 row)
> 
> But then I tried turning off the primary instance by using systemctl stop ag_test_01
> 
> PgPool recognises that the node has been shutdown with an administrative command but continues to perform a failover anyway.
> 
> Please see the logs below (the logs are for node 0 and 2, node 1 was turned off earlier and showed the same behaviour):


> 2022-07-29 08:51:13 psql [592] LOG: reading and processing packets
> 2022-07-29 08:51:13 psql [592] DETAIL: postmaster on DB node 0 was shutdown by administrative command
> 2022-07-29 08:51:13 psql [592] LOCATION: pool_process_query.c:4898

This itself did not trigger the failover, as expected.

> 2022-07-29 08:51:13 psql [592] LOG: pool_ssl: "SSL_read": "no SSL error reported"
> 2022-07-29 08:51:13 psql [592] LOCATION: pool_ssl.c:420

The failover was triggered here. Pgpool-II child process tried to
connect to node 0 but failed.  You can set failover_on_backend_error
to off to avoid the failover.

Best reagards,
--
Tatsuo Ishii
SRA OSS, Inc. Japan
English: http://www.sraoss.co.jp/index_en.php
Japanese:http://www.sraoss.co.jp