[Pgpool-general] Pgpool 3.1 errors

[Olivier NOEL]

2011/10/4 Guillaume Lelarge <guillaume at lelarge.info>:
> On Tue, 2011-10-04 at 10:39 +0200, Olivier NOEL wrote:
>> Hi,
>>
>> I compiled and installed pgpool 3.1 on Debian Wheezy yesterday. No
>> compile errors.
>>
>> But I have a LOT of errors :
>>
>> Oct  4 10:09:48 pgpool01 pgpool: 2011-10-04 10:09:48 ERROR: pid 1198:
>> Child process 7712 was terminated by segmentation fault
>> Oct  4 10:09:48 pgpool01 pgpool: 2011-10-04 10:09:48 ERROR: pid 1198:
>> Child process 8249 was terminated by segmentation fault
>> Oct  4 10:09:48 pgpool01 pgpool: 2011-10-04 10:09:48 LOG:   pid 9698:
>> do_child: failback event found. restart myself.
>> Oct  4 10:09:48 pgpool01 pgpool: 2011-10-04 10:09:48 LOG:   pid 9699:
>> do_child: failback event found. restart myself.
>> Oct  4 10:09:48 pgpool01 kernel: [30451.414053] pgpool[7712]: segfault
>> at 0 ip 00000000004495b0 sp 00007fff942728c0 error 4 in
>> pgpool[400000+dc000]
>> Oct  4 10:09:48 pgpool01 kernel: [30451.415038] pgpool[8249]: segfault
>> at 0 ip 00000000004495b0 sp 00007fff942728c0 error 4 in
>> pgpool[400000+dc000]
>> Oct  4 10:10:08 pgpool01 kernel: [30471.410705] pgpool[7880]: segfault
>> at 0 ip 00000000004495b0 sp 00007fff942728c0 error 4 in
>> pgpool[400000+dc000]
>> Oct  4 10:10:08 pgpool01 pgpool: 2011-10-04 10:10:08 ERROR: pid 1198:
>> Child process 7880 was terminated by segmentation fault
>> Oct  4 10:10:08 pgpool01 pgpool: 2011-10-04 10:10:08 LOG:   pid 9705:
>> do_child: failback event found. restart myself.
>>
>> And a LOT of :
>>
>> pid 1465: ProcessFrontendResponse: failed to read kind from frontend.
>> frontend abnormally exited
>>
>> Pgpool 3.1 on Debian Wheezy, compiled with postgresql-server-dev-9.1
>> Postgresql server 9.1 on FreeBSD 8.2 in streaming replication, 1
>> master 2 slaves (replication is working)
>>
>> Pgpool kept disconnecting Postgres backends randomly. It didn't find
>> at first the master node, I had to rename the node from
>> backend_hostname6 to backend_hostname0.
>> Another weird thing is that pgpool connects to the postgres backends
>> with "nobody" instead of "postgres".
>>
>> I can't use pgpool 3.0.4 either (version of pegpool on Debian Wheezy),
>> because I get "portal" errors :
>> ERROR:  portal "" does not exist
>> ERROR:  portal "pgpool_error_portal" does not exist
>>
>> Config file is attached.
>>
>
> The conf file you're using is a 3.0 one. I hope you don't use that for
> 3.1, as it may not work at all.

The conf file is essentially the same, no real differences, with the
exception of backend_socket_dir, which is deprecated

> Your num_init_children seems way too big. 1000... it means pgpool will
> launch 1000 subprocesses when it starts. Kinda crazy :) Moreover, with
> the max_pool at 8, it means pgpool can have 8000 connections to your
> PostgreSQL server. Have you really configured PostgreSQL with a
> max_connections of 8000?

Each Postgresql have a max_connections of 1000 (we have above 600
connections per , we planned to have 8 servers. Actually max_pool is
at 3

> On the load balancing configuration, you didn't deny load balancing of
> nextval and setval, which is obviously an error.

I use white_function_list for that, I have about 450 functions that
are DDL/DML requests, so I only list the read only functions

> On your backend configuration, you set backend_hostname0, and then go
> directly to backend_hostname7. It may be an issue. Try to set them in
> order (O, then 1, then 2). Not sure it has a real value, I never tried
> that, but it works for me when they are set in order, and without hole.

It doesn't seems to disturb pgpool, but it was having problem starting
without a node 0.

>
>
> --
> Guillaume
>  http://blog.guillaume.lelarge.info
>  http://www.dalibo.com
>
>