Re: [NOVICE] URGENT! pg_dump doesn't work! (fwd)

---------- Forwarded message ----------
Date: Tue, 23 Jul 2002 14:39:50 +0100 (BST)
From: Nigel J. Andrews <nandrews@investsystems.co.uk>
To: Wim <wdh@belbone.be>
Subject: Re: [NOVICE] [GENERAL] URGENT! pg_dump doesn't work!


On Tue, 23 Jul 2002, Wim wrote:
>
> I looked for core files in my data directory, but could find any... I
> installed gdb, since I run postgres on Solaris 9.
> I have no experience with gdb, so could you tell me how I get a trace
> when postgres fails?


What you need to do is:

1) start psql
  - to connect to your backend which I assume is running or you already know
    how to do

2) identify the backend (postgres) server your psql session from 1) is attached
to
  - use ps ax | grep postgres to list all backend processes (possibly ps -fe
    instead ps ax). If your database is quiet and only your psql session is
    using it then the relevent process is obvious. If your database is busy you
    may find it easier to connect to it as a user that is not normally used,
    for example the database superuser (usually postgres), as this information
    should appear in the process listing.

3) start gdb using:
      gdb <path to postgres backend binary> <process id from step 2>
  - this attaches gdb to process serving your psql session.

4) at the gdb prompt press 'c'
  - this makes the backend process continue normal operations (gdb stopped it
    when it attached to it)

5) in your psql session issue the statement that causes the fault

6) when the backend generates it's fault you should have a prompt in your gdb
session. Typing 'bt' at the gdb prompt will give you a stack trace.

Now, the complications.

1) To get a useful stack trace you may need to have postgres built with
debugging enabled, which would require a reconfigure and build from source if
not already done.

2) I'm starting to think I've taken you wrong route with this. I'm no expert
on the internals but I'm not convinced the backend is generating a fault such
that you would get a core file or be able to do the final stage in above
steps. My apologies for wasting your time if that's the case. However, as
you've already gone to the trouble to install gdb it may be easiest to give the
above instructions a go and see if you do indeed get a gdb prompt in the final
step. Whether this works or not you should also attempt to obtain the stack
trace from the pg_dump attempt. There should be a way to start pg_dump such
that you have time to determine the backend process for it, however, reading
the manpage I can only see -W as a possibility.

One other question that would be interesting perhaps is whether the problem you
are experiencing persists across a backend restart.

Hopefully someone like Tom or Bruce can provide some guidance. Tom already
suggested data corruption.


--
Nigel J. Andrews
Director

---
Logictree Systems Limited
Computer Consultants