[Pgpool-general] Duration of commit on more than one backend configuration

[Ramon de Carvalho Valle]

On 06/21/2010 10:32 PM, Tatsuo Ishii wrote:
>> On 06/21/2010 10:16 PM, Tatsuo Ishii wrote:
>>>> On 06/21/2010 08:31 PM, Tatsuo Ishii wrote:
>>>>>> As far as I know, pgpool-II waits for [all] backends to commit the SQL
>>>>>> statements (under replication mode) before returning to the client,
>>>>>> right? This means that if, for example, one of the backends is very busy
>>>>>> (100% CPU, etc.), pgpool will experience a delay and so will the
>>>>>> application. Is this correct?
>>>>>
>>>>> Correct. This is a known behavior of all synchronous replication
>>>>> systems. This ensures that committed data (from client's point of
>>>>> view) is actually committed to all backends.
>>>>
>>>> Is there a way to set a timeout for this? It would be interesting the
>>>> possibility of setting a timeout, and if it expires, the possibility of
>>>> automatically degenerate the backend.
>>>
>>> I think you could use PostgreSQL's statement_timeout for this purpose.
>>
>> This will automatically degenerate the backend and restore normal
>> services?
> 
> No.
> 
>> If a backend has problems, it is interesting that pgpool-II be
>> able to degenerate it and restore normal services by using the other
>> available backends.
> 
> I'm not sure just taking long time to execute a query is the trigger
> of degeneration.

This may be optional and disabled by default, but it would certainly be
very useful.

> --
> Tatsuo Ishii
> SRA OSS, Inc. Japan
> English: http://www.sraoss.co.jp/index_en.php
> Japanese: http://www.sraoss.co.jp

-- 
Ramon de Carvalho Valle
RISE Security
E-Mail: ramon at risesecurity.org