[pgpool-general: 7551] Re: Strange behavior on switchover with detach_false_primary enabled

[Tatsuo Ishii]

> On Wed, 5 May 2021, Tatsuo Ishii wrote:
> 
>>> Excellent job guys!
>>> Does this fix affect on 2-node cluster behaviour?
>>
>> I think it does not.
>>
>> If both 2-node are up, then the cluster behavior will not change
>> because the quorum exists. If one of the nodes is down,
>> detach_false_primary is meaningless anyway.
> 
> What if we have enable_consensus_with_half_votes = on ?

I thought you are talking about there are only two servers and
pgpool/PostgreSQL are running on each server. Now I realize that you
are talking about the configuration in which pgpool and PostgreSQL are
running on separate servers. Ok, let's see...

If enable_consensus_with_half_votes = on, there's a possibility of
split brain. In the worst case detach_false_primary will run on each
pgpool node. However this is same as what current detach_false_primary
is doing anyway. On the other hand, one of pgpool actually goes down,
detach_false_primary will run on the live pgpool, again which is same
as today.

If enable_consensus_with_half_votes = off, there will be difference
when one of pgpool goes down because the quorum will be lost and
detach_false_primary will not run. However if the quorum is lost, VIP
will be lost and whole DB service is lost. I don't think anyone cares
about detach_false_primary in this case.

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