> On 16 Mar 2026, at 11:24, Zsolt Parragi <[email protected]> wrote:
>
>> I'm experimenting with an ereport(FATAL_CLIENT_ONLY) option, in the
>> same vein as WARNING_CLIENT_ONLY, to try to cover this.
>
> I attached v7 that uses that and removes the abandoned handling as it
> is no longer needed with it.
>
>> P.S. I would eventually like to record our undocumented SASL profile
>> in a test suite (he said, staring at pg-pytest)...
>
> That would be definitely useful, with the todo comment and this not
> being documented I thought that this is a proper way to handle the
> issue. Even a proper documentation about it would be a good starting
> point.
> <v7-0001-Improve-OAuth-discovery-logging.patch>
I've took a look into v7. FATAL_CLIENT_ONLY approach LGTM.
pg_stat_database.sessions_fatal seems to be still incremented, but, probably,
we can live with it. But also we can fix it.
Changes to send_message_to_server_log() seems unreachable to me.
I think is_log_level_output() returns false for FATAL_CLIENT_ONLY, so
edata->output_to_server is never set to true for this level, and these
functions are never called.
FATAL_CLIENT_ONLY = 23 sits between FATAL (22) and PANIC (24).
Consider swapping FATAL and FATAL_CLIENT_ONLY, so that code like this will
have more sense:
elevel = Max(elevel, errordata[i].elevel);
Does this assignment have an effect?
+ ctx->state = OAUTH_STATE_FINISHED;
+ ereport(FATAL_CLIENT_ONLY,
Thanks!
Best regards, Andrey Borodin.
