[Pgcluster-general] Last chance to get everything working

[a.mitani at sra-europe.com]

Hi,

Yha,
Performance point of view, replication is not any help for data writing load.
On the contrary, data writing load makes very bad performance on
replication system. In that case, It becomes slower and slower as a node
is increased.

Table space, table partitioning and parallel query migh be help for
performance improvement.

Even so, server hung-up is not accepted for me.
I'll investigate cause of it.

Thank you for update me.

At.Mitani


> You might remember that I was in contact back then when we went live and
> you gave me quite a lot of support. The main problem was that servers
> kept dropping out of the cluster, recovery didn't work, and sometimes
> the whole cluster just hung. I just felt that I couldn't continue to run
> the software in a production enviroment (an hour of outtime is a 50k
> loss) and I felt that we never got to the root of the problem. You did
> make some fixes, but it never resolved our issues.
>
> I admit that our set-up maybe a bit too much high-performance - we have
> to deal with several million updates per day and maybe that's just too
> much.
>
> I think the main problem was having 4 servers in the cluster. Gut
> feeling tells me that there were timeout issues with updates - but
> that's just a guess.
>
> The environment doesn't exist any more so I'm afraid I can't do any
> further testing on this.
>
> Best Regards,
>
> Christian
>
> a.mitani at sra-europe.com wrote:
>> Hi Christian,
>>
>> It is surprise for me that the cluster hanged whithout any reason.
>> I have to correct it.
>> Would you let me know the details, if you don't mind.
>>
>> Regards,
>> ---------------------------
>> At.Mitani
>>
>>
>>> we had pgcluster running in production for about 3 weeks and then
>>> stopped as we had huge problems with the cluster just hanging without
>>> any reason - as this was driving a web site with a huge turnover, we
>>> found the performance to be unacceptable and are now back to using
>>> postgresql without replication.
>>>
>>> Regards,
>>>
>>> Christian
>>>
>>>
>>> John Gardner wrote:
>>>
>>>> I've been working with pgCluster for the last four months now, I'm at
>>>> the stage where we need to install it in a production environment, but
>>>> I
>>>> must say that I'm still apprehensive about a few things, some things
>>>> certainly don't work in my environment.  I have chatted with Rick
>>>> Vernam
>>>> about a few of these problems, and although we seem to have the same
>>>> configuration, his works and mine doesn't.
>>>>
>>>> My configuration is this:
>>>>
>>>>     DB 1               DB 2                                DB 3
>>>> +--------------+   +--------------+                  +--------------+
>>>> |   cluster_1  |   |   cluster_2  |       VPN        |   cluster_3  |
>>>> |   rep_1      |---|   rep_2      |------------------|   rep_3      |
>>>> |   lb_1       |   |   lb_2       |                  |              |
>>>> +--------------+   +--------------+                  +--------------+
>>>>        |                   |
>>>>        |                   |
>>>> +--------------+   +--------------+
>>>> |   Apache     |   |   Apache     |
>>>> |   mod_jk     |   |   mod_jk     |
>>>> |   Tomcat     |---|   Tomcat     |
>>>> +--------------+   +--------------+
>>>>      APP 1APP 2
>>>>
>>>> My problems are:
>>>>
>>>> - Replication works between all of the nodes under normal
>>>> circumstances.
>>>> - Multiple replicator failover doesn't work.
>>>> - Occasionally when bringing a DB Cluster node back up in recovery
>>>> mode,
>>>> access to the remaining DB clusters locks and I can't access anything
>>>> either via the load balancer or directly.
>>>> - Does <Response_Mode> have any use at all? One of the reasons for
>>>> going
>>>> down the pgCluster route was the 'normal' operation, but I've just
>>>> read
>>>> a post on the mailing list from Atsushi saying that normal operation
>>>> doesn't work and should be removed.
>>>> - What is the difference of setting <Use_Replication_Log> to yes or
>>>> no?
>>>>
>>>> We really, have to make a decision this week, whether we stick with
>>>> pgCluster or go via another route.  I've stuck with pgCluster for this
>>>> long because I was convinced that the reason it was not performing as
>>>> I
>>>> had expected was due to configuration errors, but now I'm not so
>>>> sure...
>>>> If anyone can give me any help I would be extremely grateful!
>>>>
>>>> Thanks in advance
>>>>
>>>> John
>>>> _______________________________________________
>>>> 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
>