Обсуждение: Error on CLUSTER command

"Mark Steben" <msteben@autorevenue.com> writes:
> I am getting the following error when I attempt to CLUSTER on a table:
>  mavmail=# cluster mailer_queue;
> ERROR:  could not create relation "pg_temp_752992924": File exists

If you repeat the command, does the error persist?  Does the number
stay the same?

> Again we are running 7.4.5.

Tsk tsk.  Why not 7.4.19, or at least some reasonably recent 7.4.x
update?

            regards, tom lane

The number may change pointing to another index or stay the same.
The error persists if I repeat 3 or 4 times.  Usually after the
4th or 5th time the cluster is successful.

What is the risk of performing an upgrade to 7.4.19?

-----Original Message-----
From: Tom Lane [mailto:tgl@sss.pgh.pa.us]
Sent: Friday, February 08, 2008 11:19 AM
To: Mark Steben
Cc: pgsql-admin@postgresql.org
Subject: Re: [ADMIN] Error on CLUSTER command

"Mark Steben" <msteben@autorevenue.com> writes:
> I am getting the following error when I attempt to CLUSTER on a table:
>  mavmail=# cluster mailer_queue;
> ERROR:  could not create relation "pg_temp_752992924": File exists

If you repeat the command, does the error persist?  Does the number
stay the same?

> Again we are running 7.4.5.

Tsk tsk.  Why not 7.4.19, or at least some reasonably recent 7.4.x
update?

            regards, tom lane

"Mark Steben" <msteben@autorevenue.com> writes:
> The number may change pointing to another index or stay the same.
> The error persists if I repeat 3 or 4 times.  Usually after the
> 4th or 5th time the cluster is successful.

Hm.  I'd guess that the installation is old enough that the OID counter
has wrapped around, and is currently in a region that is pretty densely
populated with existing tables.  (You could check that out by looking
into the database directory.)  7.4.x doesn't have the logic that was
added later to skip over filenames that are already in use; instead
you just get an error if there's a collision.

I forget whether 7.4's pg_resetxlog has an option to set the OID
counter, but if it does then a temporary fix is to move the OID counter
to some less densely populated range.

> What is the risk of performing an upgrade to 7.4.19?

Low, but it won't make this particular problem go away.

            regards, tom lane

The doc says to do an initdb and reload the data whenever you run
pg_resetxlog.  Do you concur with that even if we are only resetting the OID
counter?

-----Original Message-----
From: Tom Lane [mailto:tgl@sss.pgh.pa.us]
Sent: Friday, February 08, 2008 11:36 AM
To: Mark Steben
Cc: pgsql-admin@postgresql.org
Subject: Re: [ADMIN] Error on CLUSTER command

"Mark Steben" <msteben@autorevenue.com> writes:
> The number may change pointing to another index or stay the same.
> The error persists if I repeat 3 or 4 times.  Usually after the
> 4th or 5th time the cluster is successful.

Hm.  I'd guess that the installation is old enough that the OID counter
has wrapped around, and is currently in a region that is pretty densely
populated with existing tables.  (You could check that out by looking
into the database directory.)  7.4.x doesn't have the logic that was
added later to skip over filenames that are already in use; instead
you just get an error if there's a collision.

I forget whether 7.4's pg_resetxlog has an option to set the OID
counter, but if it does then a temporary fix is to move the OID counter
to some less densely populated range.

> What is the risk of performing an upgrade to 7.4.19?

Low, but it won't make this particular problem go away.

            regards, tom lane

On Feb 8, 2008 1:37 PM, Mark Steben <msteben@autorevenue.com> wrote:
> The doc says to do an initdb and reload the data whenever you run
> pg_resetxlog.  Do you concur with that even if we are only resetting the OID
> counter?

On 7.4 I'd say you'd be better off dumping and reloading all your data
than playing with pg_resetxlog on a live database.  Might as well
update to 7.4.19 while you're at it.

I'd run a dump restore to a test machine running 7.4.19 first to make
sure it will go well, or get another server that can handle the load,
or make a new 7.4.19 instance on the same machine and transfer to
that.  But if it's a production database I wouldn't run pg_resetxlog
on it myself.

"Mark Steben" <msteben@autorevenue.com> writes:
> The doc says to do an initdb and reload the data whenever you run
> pg_resetxlog.  Do you concur with that even if we are only resetting the OID
> counter?

The docs are thinking of the case where you are trying to recover from
database corruption.  If you just use it to change the OID counter after
a clean shutdown, I don't see a problem.

Not that it wouldn't be a good idea to take a backup first, just
in case...

            regards, tom lane