[Pgpool-general] We really need a change to the API for calling recovery

[Guillaume Lelarge]

On Thu, 2011-07-21 at 13:27 +0900, Tatsuo Ishii wrote:
> > Tatsuo,
> > 
> > Currently the recovery routine is called on the current master database,
> > and the parameters passed to it do not include the master database IP
> > address or hostname.  Given that the way to provision a node with 9.1
> > replication is via pg_basebackup, which MUST be run on the standby, this
> > makes writing a recovery script which actually works rather difficult
> > and unreliable.
> > 
> > I really think that you need to look at restructuring how the recovery
> > call for new nodes works.  What I would prefer is to have a script which
> > is called *on the pgpool server* itself, and not on the master node.
> > This script would be passed the master address and the standby address;
> > I don't think it would need any other parameters.
> 
> Maybe that is only true for streaming replication. I would think about
> how to deal with pg_basebackup. BTW, I think the script would need
> user, password and port number at least.
> 
> > If that's impossible for some reason, can we at least pass the master
> > address to the recovery script?  Working without it is unnecessarily
> > difficult.
> 
> That would be easy. I would do it for 3.1.

I thought the idea was to work on 9.1 for 3.2. Anyway, if we keep adding
things to 3.1, it will never be released.


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