Thread: BUG #3591: autovacuum crash

BUG #3591: autovacuum crash

From
"Ben Kim"
Date:
The following bug has been logged online:

Bug reference:      3591
Logged by:          Ben Kim
Email address:      bkim@tamu.edu
PostgreSQL version: 8.1.0
Operating system:   Solaris 8
Description:        autovacuum crash
Details:

The server crashed suddenly the second time since two days ago. ulimit is
unlimited, but I see no core file under the postgresql root directory.
=============================

   6780 2007-08-31 14:01:46 CDT   LOG:  autovacuum process (PID 6379) was
   terminated by signal 11
   6780 2007-08-31 14:01:46 CDT   LOG:  terminating any other active server
   processes
user db1 [local] 6375 2007-08-31 14:01:46 CDT idle 0 WARNING:  terminating
   connection because of crash of another server process
user db1 [local] 6375 2007-08-31 14:01:46 CDT idle 0 DETAIL:  The postmaster
   has commanded this server process to roll back the current transaction
and      exit, because another server process exited abnormally and possibly
corrupted   shared memory.
user db1 [local] 6375 2007-08-31 14:01:46 CDT idle 0 HINT:  In a moment you
   should be able to reconnect to the database and repeat your command.
.... (repeated for every database) ...

   6780 2007-08-31 14:01:47 CDT   LOG:  all server processes terminated;
   reinitializing
   6381 2007-08-31 14:01:47 CDT   LOG:  database system was interrupted at
2007-08-31 14:01:35 CDT
   6381 2007-08-31 14:01:47 CDT   LOG:  checkpoint record is at F/5A51F588
   6381 2007-08-31 14:01:47 CDT   LOG:  redo record is at F/5A51F588; undo
   record is at 0/0; shutdown FALSE
   6381 2007-08-31 14:01:47 CDT   LOG:  next transaction ID: 140043601; next
   OID: 25139832
   6381 2007-08-31 14:01:47 CDT   LOG:  next MultiXactId: 411; next
   MultiXactOffset: 897
   6381 2007-08-31 14:01:47 CDT   LOG:  database system was not properly
shut   down; automatic recovery in progress
   6381 2007-08-31 14:01:47 CDT   LOG:  redo starts at F/5A51F5D0
   6381 2007-08-31 14:01:47 CDT   LOG:  record with zero length at
F/5A525C18
   6381 2007-08-31 14:01:47 CDT   LOG:  redo done at F/5A525BE8
   6381 2007-08-31 14:01:47 CDT   LOG:  database system is ready
   6381 2007-08-31 14:01:47 CDT   LOG:  transaction ID wrap limit is
   1206535619, limited by database db3

Re: BUG #3591: autovacuum crash

From
Alvaro Herrera
Date:
Ben Kim wrote:
>
> The following bug has been logged online:
>
> Bug reference:      3591
> Logged by:          Ben Kim
> Email address:      bkim@tamu.edu
> PostgreSQL version: 8.1.0
> Operating system:   Solaris 8
> Description:        autovacuum crash
> Details:
>
> The server crashed suddenly the second time since two days ago. ulimit is
> unlimited, but I see no core file under the postgresql root directory.
> =============================
>
>    6780 2007-08-31 14:01:46 CDT   LOG:  autovacuum process (PID 6379) was
>    terminated by signal 11

This is a crash all right, but I think we fixed a certain number of bugs
after 8.1.0 was released.  Please install 8.1.9, and if it still
crashes, report again.  (We'll be wanting to see stack traces at that
point, so please make sure you are using a debug-enabled build, and a
non-zero coresize ulimit setting).

FWIW the core files should be stored in the data directories (PGDATA).

--
Alvaro Herrera                                http://www.CommandPrompt.com/
The PostgreSQL Company - Command Prompt, Inc.

Re: BUG #3591: autovacuum crash

From
Tom Lane
Date:
Alvaro Herrera <alvherre@commandprompt.com> writes:
> Ben Kim wrote:
>> PostgreSQL version: 8.1.0
>> The server crashed suddenly the second time since two days ago. ulimit is
>> unlimited, but I see no core file under the postgresql root directory.

> This is a crash all right, but I think we fixed a certain number of bugs
> after 8.1.0 was released.

Indeed.

> FWIW the core files should be stored in the data directories (PGDATA).

It seems likely that ulimit -c may be limited for the postmaster even if
it is not in your interactive shell --- very many systems run init
scripts with ulimit -c 0.  I'd suggest putting "ulimit -c unlimited"
right in the script that launches the postmaster.

            regards, tom lane

Re: BUG #3591: autovacuum crash

From
Ben Kim
Date:
Thanks. We upgraded the server.

On Fri, 31 Aug 2007, Alvaro Herrera wrote:

> Ben Kim wrote:
>>
>> The following bug has been logged online:
>>
>> Bug reference:      3591
>> Logged by:          Ben Kim
>> Email address:      bkim@tamu.edu
>> PostgreSQL version: 8.1.0
>> Operating system:   Solaris 8
>> Description:        autovacuum crash
>> Details:
>>
>> The server crashed suddenly the second time since two days ago. ulimit is
>> unlimited, but I see no core file under the postgresql root directory.
>> =============================
>>
>>    6780 2007-08-31 14:01:46 CDT   LOG:  autovacuum process (PID 6379) was
>>    terminated by signal 11
>
> This is a crash all right, but I think we fixed a certain number of bugs
> after 8.1.0 was released.  Please install 8.1.9, and if it still
> crashes, report again.  (We'll be wanting to see stack traces at that
> point, so please make sure you are using a debug-enabled build, and a
> non-zero coresize ulimit setting).
>
> FWIW the core files should be stored in the data directories (PGDATA).
>
> --
> Alvaro Herrera                                http://www.CommandPrompt.com/
> The PostgreSQL Company - Command Prompt, Inc.
>

Regards,

Ben K.
Developer
http://benix.tamu.edu