Обсуждение: Problem with pg_upgrade 8.3 to 9.1.4 - clog missing?!

Поиск
Список
Период
Сортировка

Problem with pg_upgrade 8.3 to 9.1.4 - clog missing?!

От
hubert depesz lubaczewski
Дата:
hi
I just upgraded test copy of database of our customer (~ 600GB of data).
upgrade went fine, no errors. but vacuumdb -azv ended with an error:

=> vacuumdb --all --analyze -p 6665
vacuumdb: vacuuming database "client_db"
vacuumdb: vacuuming database "pg_audit"
vacuumdb: vacuuming database "postgres"
vacuumdb: vacuuming of database "postgres" failed: ERROR:  could not access status of transaction 860626316
DETAIL:  Could not open file "pg_clog/0334": No such file or directory.

I know we had these kind of errors before, but I thought they all got fixed.

What can I do to debug it more?

8.3 was 8.3.16.
8.3.16 is from rpms, 9.1.4 is from source compilation.

both have integer datatimes, and both are for 64bit linux.

pg_clog contains:

-rw------- 1 postgres postgres 262144 Jun  6 21:22 04A4
-rw------- 1 postgres postgres 262144 Jun  6 21:22 04A5
-rw------- 1 postgres postgres 262144 Jun  6 21:22 04A6
-rw------- 1 postgres postgres 262144 Jun  6 21:22 04A7
-rw------- 1 postgres postgres 262144 Jun  6 21:22 04A8
-rw------- 1 postgres postgres 262144 Jun  6 21:22 04A9
-rw------- 1 postgres postgres 262144 Jun  6 21:22 04AA
-rw------- 1 postgres postgres 262144 Jun  6 21:22 04AB
-rw------- 1 postgres postgres 262144 Jun  6 21:22 04AC
-rw------- 1 postgres postgres 262144 Jun  6 21:22 04AD
-rw------- 1 postgres postgres 262144 Jun  6 21:22 04AE
-rw------- 1 postgres postgres 262144 Jun  6 21:22 04AF
-rw------- 1 postgres postgres 262144 Jun  6 21:22 04B0
-rw------- 1 postgres postgres 262144 Jun  6 21:22 04B1
-rw------- 1 postgres postgres 262144 Jun  6 21:22 04B2
-rw------- 1 postgres postgres 262144 Jun  6 21:22 04B3
-rw------- 1 postgres postgres 262144 Jun  6 21:22 04B4
-rw------- 1 postgres postgres 262144 Jun  6 21:22 04B5
-rw------- 1 postgres postgres 262144 Jun  6 21:22 04B6
-rw------- 1 postgres postgres 262144 Jun  6 21:22 04B7
-rw------- 1 postgres postgres 262144 Jun  6 21:22 04B8
-rw------- 1 postgres postgres 262144 Jun  6 21:22 04B9
-rw------- 1 postgres postgres 262144 Jun  6 21:22 04BA
-rw------- 1 postgres postgres 262144 Jun  6 21:22 04BB
-rw------- 1 postgres postgres 262144 Jun  6 21:22 04BC
-rw------- 1 postgres postgres 262144 Jun  6 21:22 04BD
-rw------- 1 postgres postgres 262144 Jun  6 21:22 04BE
-rw------- 1 postgres postgres 262144 Jun  6 21:22 04BF
-rw------- 1 postgres postgres 262144 Jun  6 21:22 04C0
-rw------- 1 postgres postgres 262144 Jun  6 21:22 04C1
-rw------- 1 postgres postgres 262144 Jun  6 21:22 04C2
-rw------- 1 postgres postgres 262144 Jun  6 21:22 04C3
-rw------- 1 postgres postgres 262144 Jun  6 21:22 04C4
-rw------- 1 postgres postgres 262144 Jun  6 21:22 04C5
-rw------- 1 postgres postgres 262144 Jun  6 21:22 04C6
-rw------- 1 postgres postgres 262144 Jun  6 21:22 04C7
-rw------- 1 postgres postgres 262144 Jun  6 21:22 04C8
-rw------- 1 postgres postgres 262144 Jun  6 21:22 04C9
-rw------- 1 postgres postgres 262144 Jun  6 21:22 04CA
-rw------- 1 postgres postgres 262144 Jun  6 21:22 04CB
-rw------- 1 postgres postgres 262144 Jun  6 21:22 04CC
-rw------- 1 postgres postgres 262144 Jun  6 21:22 04CD
-rw------- 1 postgres postgres 262144 Jun  6 21:22 04CE
-rw------- 1 postgres postgres 262144 Jun  6 21:22 04CF
-rw------- 1 postgres postgres 262144 Jun  6 21:22 04D0
-rw------- 1 postgres postgres 262144 Jun  6 21:22 04D1
-rw------- 1 postgres postgres 262144 Jun  6 21:22 04D2
-rw------- 1 postgres postgres 196608 Jun  7 00:15 04D3

unfortunately i do not have the 8.3 data dir any more.

if the problem is only with postgres database - i can ignore it. but I'd rather
be sure what is the source of the issue.

Best regards,

depesz


--
The best thing about modern society is how easy it is to avoid contact with it.
                                                             http://depesz.com/

Re: Problem with pg_upgrade 8.3 to 9.1.4 - clog missing?!

От
Bruce Momjian
Дата:
On Thu, Jun 07, 2012 at 09:16:49PM +0200, hubert depesz lubaczewski wrote:
> hi
> I just upgraded test copy of database of our customer (~ 600GB of data).
> upgrade went fine, no errors. but vacuumdb -azv ended with an error:
>
> => vacuumdb --all --analyze -p 6665
> vacuumdb: vacuuming database "client_db"
> vacuumdb: vacuuming database "pg_audit"
> vacuumdb: vacuuming database "postgres"
> vacuumdb: vacuuming of database "postgres" failed: ERROR:  could not access status of transaction 860626316
> DETAIL:  Could not open file "pg_clog/0334": No such file or directory.
>
> I know we had these kind of errors before, but I thought they all got fixed.
>
> What can I do to debug it more?
>
> 8.3 was 8.3.16.
> 8.3.16 is from rpms, 9.1.4 is from source compilation.
>
> both have integer datatimes, and both are for 64bit linux.

I assume 8.3 vacuumdb did not generate these errors.  If it was 8.4 I
would suggest it was because we don't copy visibility map files from
pre-9.1 because those were not crash-safe, but 8.3 didn't have
visibility map files (added in PG 8.4), so 8.3 should have been checking
the exact same transaction ids as 9.1.

Basically, I can't think of a cause.

--
  Bruce Momjian  <bruce@momjian.us>        http://momjian.us
  EnterpriseDB                             http://enterprisedb.com

  + It's impossible for everything to be true. +

Re: Problem with pg_upgrade 8.3 to 9.1.4 - clog missing?!

От
hubert depesz lubaczewski
Дата:
On Wed, Jun 13, 2012 at 02:32:21PM -0400, Bruce Momjian wrote:
> I assume 8.3 vacuumdb did not generate these errors.  If it was 8.4 I
> would suggest it was because we don't copy visibility map files from
> pre-9.1 because those were not crash-safe, but 8.3 didn't have
> visibility map files (added in PG 8.4), so 8.3 should have been checking
> the exact same transaction ids as 9.1.
>
> Basically, I can't think of a cause.

yeah, neither can we. but - any other test - even using the same source
of $PGDATA went flawless. I'm tempted to assume it's just a random
glitch in matrix, and just go on.

Best regards,

depesz

--
The best thing about modern society is how easy it is to avoid contact with it.
                                                             http://depesz.com/

Re: Problem with pg_upgrade 8.3 to 9.1.4 - clog missing?!

От
Bruce Momjian
Дата:
On Wed, Jun 13, 2012 at 08:43:23PM +0200, hubert depesz lubaczewski wrote:
> On Wed, Jun 13, 2012 at 02:32:21PM -0400, Bruce Momjian wrote:
> > I assume 8.3 vacuumdb did not generate these errors.  If it was 8.4 I
> > would suggest it was because we don't copy visibility map files from
> > pre-9.1 because those were not crash-safe, but 8.3 didn't have
> > visibility map files (added in PG 8.4), so 8.3 should have been checking
> > the exact same transaction ids as 9.1.
> >
> > Basically, I can't think of a cause.
>
> yeah, neither can we. but - any other test - even using the same source
> of $PGDATA went flawless. I'm tempted to assume it's just a random
> glitch in matrix, and just go on.

Is the vacuumdb still repeatedly failing?

--
  Bruce Momjian  <bruce@momjian.us>        http://momjian.us
  EnterpriseDB                             http://enterprisedb.com

  + It's impossible for everything to be true. +

Re: Problem with pg_upgrade 8.3 to 9.1.4 - clog missing?!

От
hubert depesz lubaczewski
Дата:
On Wed, Jun 13, 2012 at 02:55:41PM -0400, Bruce Momjian wrote:
> On Wed, Jun 13, 2012 at 08:43:23PM +0200, hubert depesz lubaczewski wrote:
> > On Wed, Jun 13, 2012 at 02:32:21PM -0400, Bruce Momjian wrote:
> > > I assume 8.3 vacuumdb did not generate these errors.  If it was 8.4 I
> > > would suggest it was because we don't copy visibility map files from
> > > pre-9.1 because those were not crash-safe, but 8.3 didn't have
> > > visibility map files (added in PG 8.4), so 8.3 should have been checking
> > > the exact same transaction ids as 9.1.
> > >
> > > Basically, I can't think of a cause.
> >
> > yeah, neither can we. but - any other test - even using the same source
> > of $PGDATA went flawless. I'm tempted to assume it's just a random
> > glitch in matrix, and just go on.
>
> Is the vacuumdb still repeatedly failing?

well - in this particular upgraded database - i think so. I did run it
couple of times, it failed all the times.
But this $PGDATA is not longer used - we did another tests, and no other
test showed problems - i.e. vacuum worked fine.

Best regards,

depesz