Обсуждение: Vacuum doesn't end

I have an application that may add a couple million rows per day so I
vacuum nightly.  The tables never get to more than about 10 million rows
before I move off the interesting information to other media.  Somewhat
often, my vacuums don't complete as shown from this ps command (usually
VACUUMs take a couple minutes)

4442 ?        R    949:07 postgres: postgres tag 127.0.0.1(33420) VACUUM

and the CPU for postmaster (hyperthreaded linux) will be in the high
90s.

The locks look like:
scouts=# select * from pg_locks;
 relation | database | transaction |  pid  |           mode           |
granted
----------+----------+-------------+-------+--------------------------+---------
          |          |    12826125 | 11642 | ExclusiveLock            |
t
    16839 |    17230 |             | 11642 | AccessShareLock          |
t
    17251 |    17230 |             |  4442 | RowExclusiveLock         |
t
    17251 |    17230 |             |  4442 | ShareUpdateExclusiveLock |
t
    17246 |    17230 |             |  4442 | ShareUpdateExclusiveLock |
t
    17246 |    17230 |             |  4442 | ShareUpdateExclusiveLock |
t
          |          |    12817402 |  4442 | ExclusiveLock            |
t
(7 rows)


I all inserts and maintenance through JDBC and may have inserts going on
while a different java thread calls the VACUUM command.

Any thoughts?  Can I recover without dropping the server?

Thanks for any help,
Scott

Usually, on this situations, you've got a deadlock or alike.

What I do is to trace the pid of the vacuum (strace -p pid) and try to guess if the vacuum
is working or expecting some semaphore to get green (that would be a problem on the db side).

If this is the problem, a quick look into the procs running in the pg_Stat_activity table will give
you an idea of the situation. Prolly, nothing was going on when you started your vacuum
but somehow, a proc acquired a lock somewhere and the vacuum reached that somewhere causing
somekinda wait.

At least, this is very often in my scenery, using 7.4.3  and linux also.

Regards,
Guido




--
"Adopting the position that you are smarter than an automatic
optimization algorithm is generally a good way to achieve less
performance, not more" - Tom Lane.

strace did show what looked like it was stuck waiting, and it was the
same vacuum proc as in the pg_Stat_activity.  I had to restart the
postmaster and eventually reload the data because no subsequent vacuum
would finish properly.  Was there anything else I could have done?

On that note, if I am just filling tables, extracting the data off,
dropping the tables, and then recreating additional tables, do I really
need to do a VACUUM?

Thanks,
Scott



On Wed, 2005-08-03 at 09:41, Guido Barosio wrote:
> Usually, on this situations, you've got a deadlock or alike.
>
> What I do is to trace the pid of the vacuum (strace -p pid) and try to
> guess if the vacuum
> is working or expecting some semaphore to get green (that would be a
> problem on the db side).
>
> If this is the problem, a quick look into the procs running in the
> pg_Stat_activity table will give
> you an idea of the situation. Prolly, nothing was going on when you
> started your vacuum
> but somehow, a proc acquired a lock somewhere and the vacuum reached
> that somewhere causing
> somekinda wait.
>
> At least, this is very often in my scenery, using 7.4.3  and linux
> also.
>
> Regards,
> Guido
>
> On 8/3/05, Scott Barvick <sbarvick@comcast.net> wrote:
>         I have an application that may add a couple million rows per
>         day so I
>         vacuum nightly.  The tables never get to more than about 10
>         million rows
>         before I move off the interesting information to other
>         media.  Somewhat
>         often, my vacuums don't complete as shown from this ps command
>         (usually
>         VACUUMs take a couple minutes)
>
>         4442 ?        R    949:07 postgres: postgres tag
>         127.0.0.1(33420) VACUUM
>
>         and the CPU for postmaster (hyperthreaded linux) will be in
>         the high
>         90s.
>
>         The locks look like:
>         scouts=# select * from pg_locks;
>         relation | database | transaction |  pid  |
>         mode           |
>         granted
>         ----------+----------+-------------+-------+--------------------------+---------
>                   |          |    12826125 | 11642 |
>         ExclusiveLock            |
>         t
>             16839 |    17230 |             | 11642 |
>         AccessShareLock          |
>         t
>             17251 |    17230 |             |  4442 |
>         RowExclusiveLock         |
>         t
>             17251 |    17230 |             |  4442 |
>         ShareUpdateExclusiveLock |
>         t
>             17246 |    17230 |             |  4442 |
>         ShareUpdateExclusiveLock |
>         t
>             17246 |    17230 |             |  4442 |
>         ShareUpdateExclusiveLock |
>         t
>                   |          |    12817402 |  4442 |
>         ExclusiveLock            |
>         t
>         (7 rows)
>
>
>         I all inserts and maintenance through JDBC and may have
>         inserts going on
>         while a different java thread calls the VACUUM command.
>
>         Any thoughts?  Can I recover without dropping the server?
>
>         Thanks for any help,
>         Scott
>
>
>
>
>         ---------------------------(end of
>         broadcast)---------------------------
>         TIP 1: if posting/reading through Usenet, please send an
>         appropriate
>                subscribe-nomail command to majordomo@postgresql.org so
>         that your
>                message can get through to the mailing list cleanly
>
>
>
> --
> "Adopting the position that you are smarter than an automatic
> optimization algorithm is generally a good way to achieve less
> performance, not more" - Tom Lane.