[pgpool-hackers: 2244] Re: WIP patch for 0000289: Inconsistent backend state

[Tatsuo Ishii]

Usama,

I have looked through patch and it seems good to me.

One thing I am not sure is how watchdog handles the case when new
backend node is added. I think it is required to add the new node info
to each pgpool.conf by hand. Does your patch improve that in this
regard?

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

> Hi Ishii-San
> 
> Can you please have a look at the attached patch to sync backend nodes
> states over the watchdog when the Pgpool-II nodes joins the watchdog
> cluster as a standby node.
> 
> Currently the Pgpool-II syncs the backend node states only at the time of
> startup which works fine for almost all cases except when the watchdog
> cluster becomes partitioned (because of some network problem) and after
> recovering from it when the Pgpool-II nodes (that are already up and
> serving) joins back the cluster. At that time the backend node status among
> different nodes can become In-sync among Pgpool-II nodes, which can be a
> major problem. The same type of scenario is also mentioned in bug report of
> 0000289: Inconsistent backend state (
> http://www.pgpool.net/mantisbt/view.php?id=289)
> 
> This work in progress patch solves this by syncing the backend status from
> the master watchdog node every time the Pgpool-II node's state change back
> to STANDBY. And the tricky part in this is how to service the pgpool-II
> children processes after updating the backend node status.
> 
> So can you please review the attached patch more specifically for the
> following two changes.
> 
> 1-) The *sync_backend_from_watchdog**()* function logic where it decides
> when to restart the pgpool-II child process. (The most of the logic is
> adopted from the failover() function)
> 
> 2-) The multiplexing of SIGUSR1 signal in pgpool-II main process. Now
> SIGUSR1 is used for two types of interrupts. a) failover and b) watchdog
> state change intimations.
> 
> Please note that the patch is generated against master branch and is still
> work in progress, So it has little extra and elevated log messages, which I
> will adjust before after completing the testing.
> 
> 
> Thanks
> Best Regards
> Muhammad Usama