Re: process type escape for log_line_prefix

From: Andres Freund <andres(at)anarazel(dot)de>
To: Jeff Janes <jeff(dot)janes(at)gmail(dot)com>
Cc: Christoph Berg <myon(at)debian(dot)org>, Michael Paquier <michael(dot)paquier(at)gmail(dot)com>, PostgreSQL mailing lists <pgsql-hackers(at)postgresql(dot)org>
Subject: Re: process type escape for log_line_prefix
Date: 2016-10-17 20:35:51
Message-ID: 20161017203551.ak42tydrfc5fkk6b@alap3.anarazel.de
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-hackers

On 2016-10-15 17:43:40 -0700, Jeff Janes wrote:
> On Fri, Oct 14, 2016 at 11:51 AM, Andres Freund <andres(at)anarazel(dot)de> wrote:
>
> > On 2016-10-14 13:11:51 +0200, Christoph Berg wrote:
> > > Re: Michael Paquier 2016-02-10 <CAB7nPqS=wBbZzBcty1KyN-
> > 5Y9bPXZ+deJbfcCtebf06eF2Uyvg(at)mail(dot)gmail(dot)com>
> > > > On Mon, Feb 8, 2016 at 11:32 PM, Andres Freund <andres(at)anarazel(dot)de>
> > wrote:
> > > > > Frequently when reading postgres logs to do some post mortem analysis
> > > > > I'm left wondering what process emitted an error/log message. After
> > the
> > > > > fact it's often hard to know wether an error message was emitted by a
> > > > > user backend or by something internal, say the checkpointer or
> > > > > autovacuum. Logging all process starts is often impractical given
> > the
> > > > > log volume that causes.
> > > > >
> > > > > So I'm proposing adding an escape displaying the process title (say
> > 'k'
> > > > > for kind?). So %k would emit something like "autovacuum worker
> > process",
> > > > > "wal sender process" etc.
> > > >
> > > > It would be nice to get something consistent between the ps output and
> > > > this new prefix, say with for example a miscadmin.h parameter like
> > > > MyProcName.
> > > >
> > > > > I'm thinking it might make sense to give normal connections "" as the
> > > > > name, they're usually already discernible.
> > > >
> > > > Yeah, that sounds fine for me. What about background workers? I would
> > > > think that they should use BackgroundWorker->bgw_name.
> > >
> > > (Rediscovering an old horse)
> > >
> > > Couldn't these processes just set %a = application_name?
> >
> > It'd not get me what I'd want, no. E.g for walsenders that'd not be
> > parsable in a meaningful way. I really would like an escape that'd
> > always output one of:
> > Postmaster, Startup, BgWriter, Checkpointer, WalWriter, WalReceiver,
> > AutovacLauncher, AutovacWorker, PgArch, PgStat, SysLogger, Backend,
> > BackgroundWorker.
> >
>
> I'm not sure what you are proposing. Which of those 13 strings you listed
> would a walsender advertise itself as?

Oops, left that one out (as it's not one of the pids explicitly listed
in postmaster.c, which I went over). 'WalSender'.

> Why would stuffing one of those words into %k be different than
> stuffing that same word into %a, where %a would otherwise be the empty
> string?

Because you very well might want to keep tracking application_name for
walsenders - as that's important e.g. for sync replica tracking purposes
- and be able to categorize log messages by the type of process.

- Andres

In response to

Browse pgsql-hackers by date

  From Date Subject
Next Message Jim Nasby 2016-10-17 20:39:04 Re: bit|varbit #, xor operator
Previous Message Tom Lane 2016-10-17 20:35:11 Re: Re: [COMMITTERS] pgsql: Replace PostmasterRandom() with a stronger way of generating ran