Обсуждение: Postgres won't start

On 08/08/2013 03:02 PM, Oliver Elphick wrote:
> Linux Mint (from Ubuntu) version 9.1.
>
> Postgres will no longer start, but I cannot find out why.

So anything happen between the last time it started and now?:

Upgrade of Postgres?
Upgrade of Mint?
Something else?

>
> Command line:
> $ /usr/lib/postgresql/9.1/bin/pg_ctl start -D /home/postgresql/9.1/main
> -l /var/log/postgresql/postgresql-9.1-main.log -s -w -o '-c
> config_file="/etc/postgresql/9.1/main/postgresql.conf"'
> pg_ctl: could not start server


--
Adrian Klaver
adrian.klaver@gmail.com

On 08/08/2013 03:17 PM, Oliver Elphick wrote:
> I tried to change the listen_addresses line in postgresql.conf, by
> adding an IPv6 address. On meeting problems I tried changing it back.

What problems?

Have you run ps to see if there is another instance of Postgres running?

>
> Currently it says:
>
> listen_addresses = 'localhost'
>


--
Adrian Klaver
adrian.klaver@gmail.com

On 08/08/2013 04:02 PM, Oliver Elphick wrote:
> To start with, it worked but the pg_hba.conf entry appeared to be wrong.
>   I tried changing that and then the current problem started.  I tried
> "listen_addresses = '*'"; then back to just 'localhost'.
>
> Since I have maximum logging enabled, I don't think it is getting as far
> as reading the configuration files - that is not mentioned in the log.
>
> There is no other instance of postgres running.

Hmmm. Two things, then I am tapped out for the time being:

1) Did you change the pg_hba.conf entry back?

2) What user are you running the pg_ctl command as?

>
>
>


--
Adrian Klaver
adrian.klaver@gmail.com

Oliver Elphick <olly@lfix.co.uk> writes:
> Linux Mint (from Ubuntu) version 9.1.
> Postgres will no longer start, but I cannot find out why.

> Command line:
> $ /usr/lib/postgresql/9.1/bin/pg_ctl start -D /home/postgresql/9.1/main -l
> /var/log/postgresql/postgresql-9.1-main.log -s -w -o '-c
> config_file="/etc/postgresql/9.1/main/postgresql.conf"'
> pg_ctl: could not start server
> Examine the log output.

> There is no error message in the log:

This log file seems suspiciously incomplete.  Is it the file mentioned in
the pg_ctl -l switch?  If so, that file would only be used until the
syslogger process took over logging (which we know it did since we see its
exit messages in there).  To find out what happened after that, you need
to look wherever your postgresql.conf is telling PG to log.

I wonder whether we shouldn't change the syslogger to emit something to
stderr when it takes over logging, saying "logging is now redirected to
<someplace>".  This isn't the first case I've seen of people forgetting
to look in the right place.

            regards, tom lane

Since I have maximum logging enabled, I don't think it is getting as far as reading the configuration files - that is not mentioned in the log.

Oliver Elphick <olly@lfix.co.uk> writes:
> On 9 August 2013 02:49, Tom Lane <tgl@sss.pgh.pa.us> wrote:
>> I wonder whether we shouldn't change the syslogger to emit something to
>> stderr when it takes over logging, saying "logging is now redirected to
>> <someplace>".

> Shouldn't you also, or instead, log to stderr just before leaving it, in
> case the configuration of the new logging is faulty?

I think the faulty-config case is (mostly?) covered already, in that we
check validity of GUC values before installing them.  The point of this
proposal is just to remind people that the "original" log file might not
be all the log output there is.  The cases that I've seen seemed to
usually involve package installations where people weren't really aware
of the nondefault logging choices that the packager had made.

Anyway, there's a concrete patch proposal in pgsql-hackers, at
http://www.postgresql.org/message-id/11372.1376015537@sss.pgh.pa.us
Please comment on that if you want to discuss details.

            regards, tom lane

... btw, were you able to resolve your original problem?  What was it?

            regards, tom lane