From: | Andrew Dunstan <andrew(at)dunslane(dot)net> |
---|---|
To: | PostgreSQL-development <pgsql-hackers(at)postgresql(dot)org> |
Subject: | logging blemishes |
Date: | 2005-09-19 23:32:25 |
Message-ID: | 432F4A89.9020103@dunslane.net |
Views: | Raw Message | Whole Thread | Download mbox | Resend email |
Thread: | |
Lists: | pgsql-hackers |
While preparing for a presentation, I noticed some mildly ugly effects
with log_line_prefix during session startup if log_connections is turned on.
Example (log_line_prefix = '%t %q%u(at)%d %r %p %c:%l'):
2005-09-19 19:16:39 EDT [unknown](at)[unknown] 6541 432f46d7.198d:1 LOG:
connection received: host=[local] port=
2005-09-19 19:16:42 EDT [unknown](at)[unknown] 6543 432f46da.198f:1 LOG:
connection received: host=[local] port=
2005-09-19 19:16:42 EDT postgres(at)template1 [local] 6543 432f46da.198f:2
LOG: connection authorized: user=postgres database=template1
2005-09-19 19:16:44 EDT postgres(at)template1 [local] 6543 432f46da.198f:3
LOG: disconnection: session time: 0:00:01.95 user=postgres
database=template1 host=[local] port=
I'm also unclear why we get two lines at the start of each connection -
that seems like a bug, or is it due to some sort of protocol
negotiation. This is a fresh FC4 machine with only pg 8.0.3 ever installed.
Anyway, currently, we test for "stop producing output here" with the
following code in elog.c:
case 'q':
/* in postmaster and friends, stop if %q is seen */
/* in a backend, just ignore */
if (MyProcPort == NULL)
i = format_len;
break;
I'm wondering if we should extend that test slightly, to something like
if (MyProcPort == NULL || MyProcPort->username == NULL
|| *(MyProcPort->username) == '\0')
Thoughts?
cheers
andrew
From | Date | Subject | |
---|---|---|---|
Next Message | Mark Dilger | 2005-09-20 00:47:39 | Re: 64-bit API for large objects |
Previous Message | Bruce Momjian | 2005-09-19 22:35:42 | Re: Spinlocks, yet again: analysis and proposed patches |