Horton

Re: Multiple hosts in connection string failed to failover in non-hot standby mode

Michael Paquier <michael@paquier.xyz>

View thread View on postgr.es
From: Michael Paquier <michael@paquier.xyz>
To: Tom Lane <tgl@sss.pgh.pa.us>
Cc: Justin Pryzby <pryzby@telsasoft.com>, Hubert Zhang <zhubert@vmware.com>, tsunakawa.takay@fujitsu.com, pgsql-hackers@postgresql.org, Andreas Seltenreich <seltenreich@gmx.de>, Bruce Momjian <bruce@momjian.us>
Date: 2021-06-01T07:34:44Z
Lists: pgsql-hackers

Attachments

On Mon, May 31, 2021 at 09:07:38PM -0400, Tom Lane wrote:
> Michael Paquier <michael@paquier.xyz> writes:
>>> Agreed, that seems bogus.
> 
>> There may be others, and I have not checked yet.  I'd rather do a
>> backpatch for this part, would you agree?
> 
> +1

Playing with all those variables and broken values here and there, I
have been able to break a bunch of tests.  Most of the failures were
in the authentication and SSL tests, but there were also fancier
cases.  For example, PGCLIENTENCODING would cause a failure with
pg_ctl, for any TAP test.

I got surprised that enforcing values for most of the PGSSL* ones did
not cause a failure when it came to the certs, CRLs keys and root
certs now.  Still, I think that we'd be safer to cancel these as
well.

Attached is the list I am finishing with.  I'd like to fix that, so
please let me know if there are any comments or objections.
--
Michael

Commits

  1. Clear conn->errorMessage at successful completion of PQconnectdb().

  2. Avoid ECPG test failures in some GSS-capable environments.

  3. Try next host after a "cannot connect now" failure.

  4. Uniformly identify the target host in libpq connection failure reports.

  5. Allow pg_regress.c wrappers to postprocess test result files.

  6. In libpq, always append new error messages to conn->errorMessage.