Thread: Postmaster failing to start on reboot

surabhi.ahuja wrote:
> Hello everyone,
>
> I reboot my machine while postmaster is up.
>
> after the m/c gets rebooted , I grep for the process "postmaster",
> and it does not find it.
>
> So,  I try starting postmaster. and it displays the following error
> message: HINT:  If you're sure there are no old server processes
> still running, remove the shared memory block with the command "ipcr
> m", or just delete the file
> "/export/home1/sdc_image_pool/dbx/postmaster.pid".
>
> does it means that i will have to delete the postmaster.pid file ..in
> such a scenarion always?

No, only when it doesn't close down properly. Check your system logs to
see what happened.

--
   Richard Huxton
   Archonet Ltd

Richard Huxton <dev@archonet.com> writes:
> surabhi.ahuja wrote:
>> So,  I try starting postmaster. and it displays the following error
>> message: HINT:  If you're sure there are no old server processes
>> still running, remove the shared memory block with the command "ipcr
>> m", or just delete the file
>> "/export/home1/sdc_image_pool/dbx/postmaster.pid".
>>
>> does it means that i will have to delete the postmaster.pid file ..in
>> such a scenarion always?

> No, only when it doesn't close down properly. Check your system logs to
> see what happened.

Also, what PG version is this exactly?  More recent versions have better
defenses against being fooled by stale postmaster.pid files.  It matters
what postmaster startup script you're using, too.

            regards, tom lane

surabhi.ahuja wrote:
> i am using PostgreSQL 8.0.0

You should upgrade to 8.0.4 as soon as is convenient - there are 4 sets
of bugfixes available.

> and the statrtup script i am using is as follows:
>
>
> *****
> #! /bin/sh
> # dbxd    Script for starting up the PostgreSQL
> #               server in the daemon mode
> #
> #
>
> # postgreSQL version is:
> PGVERSION=8.0
> # Name of the script
> NAME=dbxd
> # Command issued start/stop/restart
> action="$1"
> # Get SDC configiration
> #   . $SDCHOME/.SdCrc

I don't recognise this script. Can I ask two questions:

Q1. What distribution of Linux are you running?
Q2. How did you install PostgreSQL?
Q3. Have you made any changes to this script?

In another email you mention that this script sometimes doesn't stop PG.
This is the relevant block of code, and you can see that the line
starting "su -l postgres" has been commented out and replaced.

That's strange, because my copy of pg_ctl refuses to run as root.

> stop(){
>         echo "Stopping ${NAME} service: "
>         if [ "`uname`" = "Linux" ]; then
>            #su -l postgres -s /bin/sh -c "$PGCTL stop -D $PGDATA -s -m fast" > /dev/null 2>&1
>            /bin/sh -c "$PGCTL stop -D $PGDATA -s -m fast" > /dev/null 2>&1
>         fi
>        ret=$?
>         if [ $ret -eq 0 ]
>         then
>                 #echo "success"
>                 echo_success
>         else
>                 echo_failure
>                 if [ "`uname`" = "Linux" ]; then
>                    #su -l postgres -s /bin/sh -c "$PGCTL stop -D $PGDATA -s -m immediate" > /dev/null 2>&1
>                    /bin/sh -c "$PGCTL stop -D $PGDATA -s -m immediate" > /dev/null 2>&1
>                 fi
>         fi
>         echo
> }


> Please see the highlighted section (in red) .

Most people will be viewing this in text-mode (including me). They won't
see the red. It appears to be deleting the unix socket and lock *AFTER*
starting up PG. I don't understand how that makes sense.

Q4. Where did you get this script?

 > do i also need to remove the pid file also .

What, after starting PG? How could that make sense?

 > i mean with the other rm commands do i need to give
 > rm -f $PGDATA /postmaster.pid also?

The standard script should start and stop PostgreSQL quite successfully.
If it isn't working the first place to look is your logs.
--
   Richard Huxton
   Archonet Ltd

Richard Huxton <dev@archonet.com> writes:
> In another email you mention that this script sometimes doesn't stop PG.
> This is the relevant block of code, and you can see that the line
> starting "su -l postgres" has been commented out and replaced.
> That's strange, because my copy of pg_ctl refuses to run as root.

We added the no-root check to pg_ctl in 8.0beta4, so AFAICS this script
should *never* work to stop the postmaster if you are indeed dealing
with an 8.0.0 release.  The script change must have been made by someone
who was dealing with a pre-8.0 version (and even then, not one of the
later subreleases, because every branch since Oct 2004 has had this
check).

At least, it'd never work if executed by root.  It might work if
executed by postgres --- dunno whether any of the other stuff is
root-only.

In any case, the short answer is to get rid of this script and get
one that works properly ...

            regards, tom lane