[Pgcluster-general] error log

[Christian Dannemann]

Hi Atsushi,

I've applied the patch - but I think my problem is the replication.

What happens is that I can not issue updates to cluster4 any more - 
cluster1 - 3 are fine, but cluster4 just hangs there... I guess that's 
why the load balancer drops the server, which is correct behaviour.

The problem is finding out what happens on cluster4....

I'm running the cluster on three servers atm.

Best Regards,

Christian

a.mitani at sra-europe.com wrote:
> Hi Christian,
>
> Huum, it seems a problem of loadbalancer.
>
> Would you adapt following patch to "src/pgcluster/pglb/child.c"
>
> ex.
> $ cd src/pgcluster/pglb
> $ patch -p1 < this.patch
>
> ---------- patch for child.c--------
> --- child.c 2007-01-09 14:08:21.000000000 +0100
> +++ child.c.new 2007-02-14 16:00:30.050051794 +0100
> @@ -377,7 +377,7 @@
>             /* fatal error occured. just exit myself... */
>             case POOL_FATAL:
>                 show_error("%s:do_child: fatal error occured",func);
> -               notice_backend_error();
> +               /* notice_backend_error(); */
>                 break;
>
>             /* not implemented yet */
> ---------- end of patch -----------
>
> Regards,
> -------------
> At.Mitani
>
>   
>> Hi Atsushi,
>>
>> Yes, I'm starting with the -l option, but I don't have any errors in the
>> log when my cluster server 4 falls over.
>>
>> Strangely enough, I only have problems with one server, the rest never
>> had a failure.
>>
>> Nothing in the logs, only pglb.sts reports that the server is down and I
>> can't issue updates to it any more - it just hangs.
>>
>> Any help appreciated.
>>
>> Thanks,
>>
>> Christian
>>
>>
>> a.mitani at sra-europe.com wrote:
>>     
>>> Hi Christian,
>>>
>>> Did you use "-l" without "-v" option ?
>>> When set "-v" option, replication server reports debug messages (it's
>>> too
>>> much as you know).
>>> However, "-l" option is reporting only error messages without debug
>>> messages.
>>>
>>> Is it some help for you ?
>>> ----------------------------
>>> At.Mitani
>>>
>>>
>>>       
>>>> Hi All,
>>>>
>>>> We are having problems with our four server cluster. At the moment, we
>>>> process about a million updates per day (and a couple of million
>>>> selects). Once in a while (e.g. every couple of days), the cluster dies
>>>> - most of the time, one of the servers of the cluster is reported
>>>> faulty
>>>> by the load balancer and is then not used any more.
>>>>
>>>> As we process so many updates, we can't run asterisk in verbose mode as
>>>> the full logging slows the cluster down to a rate where we can't use it
>>>> any more.
>>>>
>>>> Would it be possible to make a small change that will log errors to the
>>>> pgreplicate.log even when not run verbose? Could the same change be
>>>> made
>>>> for the load balancer so we can find out why it suddenly drops a
>>>> server?
>>>>
>>>> Kind Regards,
>>>>
>>>> Christian
>>>> _______________________________________________
>>>> Pgcluster-general mailing list
>>>> Pgcluster-general at pgfoundry.org
>>>> http://pgfoundry.org/mailman/listinfo/pgcluster-general
>>>>
>>>>
>>>>         
>>> _______________________________________________
>>> Pgcluster-general mailing list
>>> Pgcluster-general at pgfoundry.org
>>> http://pgfoundry.org/mailman/listinfo/pgcluster-general
>>>
>>>       
>> _______________________________________________
>> Pgcluster-general mailing list
>> Pgcluster-general at pgfoundry.org
>> http://pgfoundry.org/mailman/listinfo/pgcluster-general
>>
>>     
>
> _______________________________________________
> Pgcluster-general mailing list
> Pgcluster-general at pgfoundry.org
> http://pgfoundry.org/mailman/listinfo/pgcluster-general
>   

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://pgfoundry.org/pipermail/pgcluster-general/attachments/20070214/29133fc9/attachment.html