[Pgpool-hackers] proposal for new feature

[Guillaume Lelarge]

On Fri, 2011-11-18 at 21:15 +0900, mitani at sraw.co.jp wrote:
> Hi Ishii-San,
> 
> Thank you for quick response.
> 
> In order to implement of redundancy for pgpoll, 
> I think three functions are required as follows,
> Observation, Take over and Recovery.
> 
> How to share the information, 
>  that's a key point of this solution.
> There are two methods to share the information.
>  (1) servers use a data at same data location,
>  (2) servers use same contents of data in own storage.
> In either case have some pros and cons.
> 
> I'm planing to use the (1) method cause it can make simple.
> And I'm going to use database as a shared data repository for pgpools.
> In that case, the communication protocol also might be based on the libpq.
> 
> Anyway, this is still just an idea.
> I'd like to have any comments for it.
> 

Indeed, it's really interesting. I'm only wondering what kind of
informations you'll need to keep. I would rather keep the multimaster
capacity of pgpool like you said in your previous mail. So I guess the
informations needed are the nodes' status, and the current action on a
node. Is there anything more?


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