Re: Fw: [BUGS] BUG #6011: Some extra messages are output in the event log at PostgreSQL startup

From: "Kevin Grittner" <Kevin(dot)Grittner(at)wicourts(dot)gov>
To: "MauMau" <maumau307(at)gmail(dot)com>, "Robert Haas" <robertmhaas(at)gmail(dot)com>
Cc: <pgsql-hackers(at)postgresql(dot)org>
Subject: Re: Fw: [BUGS] BUG #6011: Some extra messages are output in the event log at PostgreSQL startup
Date: 2011-05-13 15:09:43
Message-ID: 4DCD0367020000250003D6B4@gw.wicourts.gov
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-hackers

"MauMau" <maumau307(at)gmail(dot)com> wrote:
> From: "Robert Haas" <robertmhaas(at)gmail(dot)com>

>> I think Tom had the right idea upthread: what we should do is
>> make the "-s" option to pg_ctl suppress these messages (as it
>> does with similar messages on Linux). Removing them altogether
>> seems like overkill, for the reasons you mention.

Agreed.

> So I'll wait for more opinions for a week, then write and test a
> patch, and submit it as a reply to this mail thread.

Sounds reasonable.

> Treat it as a bug. (I hope some committer will kindly back-patch
> to older versions.)
>
> Make pg_ctl's -s option suppress informational event logging.

This will ultimately be up to a committer (and I'm not one), but to
me it seems reasonable to back-patch if it is addressed this way.

> Existing software which use PostgreSQL needs to be modified to add
> -s to pg_ctl. Moreover, pg_ctl unregister and pg_ctl register must
> be performed to make the patch effective in existing
> installations.

That is probably a very *good* thing if you want it to be considered
for back-patch. Having a bug fix release arbitrarily change
existing behavior which isn't data-destroying or a security risk is
not a good thing and is contrary to PostgreSQL policy for
maintaining stable branches. We can't know who might be, for
example, pulling such messages out of their logs for reporting or
monitoring purposes. If we made changes that can conceivably break
things on applying bug fix releases, we would have fewer people
applying them, and that would be bad for everyone.

> The two messages in question may be just annoying to users, and
> they might want those messages to disappear without -s. They claim
> that it is inconsistent that those messages are not recorded in
> syslog on UNIX/Linux.

I can only dream of what it's like to work somewhere that fussing
over two informational log messages on an infrequent event like
restarting a database (that *is* an infrequent event, right?) would
be something I had time for. They are very fortunate people to be
in such a position. It would appear that finding the time to add
the -s switch shouldn't be too hard in such an environment.

> the PostgreSQL Windows service must be registered by "pg_ctl
> register -s" to make use of this patch. However, according to the
> current manual, "pg_ctl register" does not take -s option.
> Actually, pg_ctl does not refuse to take -s, so this is not a big
> problem.
>
> pg_ctl register [-N servicename] [-U username] [-P password]
> [-D datadir] [-w] [-t seconds] [-o options]

When you write the patch, be sure to include a fix for the docs
here, please.

Thanks for taking the time to work through the issue.

-Kevin

In response to

Responses

Browse pgsql-hackers by date

  From Date Subject
Next Message Robert Haas 2011-05-13 15:18:04 Re: Unix latch implementation that wakes on postmaster death
Previous Message Tom Lane 2011-05-13 14:48:01 Re: Unix latch implementation that wakes on postmaster death