[Pgpool-hackers] path to follow master after a failover

[Gilles Darold]

Le 02/02/2011 00:33, Tatsuo Ishii a écrit :
>>> - IMO we need finer control over which node should be
>>>   degenerated. Probably we should have a new flag for each
>>>   backend something like this:
>>>
>>>   backend_option0 = opt_value where opt_value is one of:
>>>
>>>   DEGENRATE_IF_NEEDED: degenrate whenever pgpool-II thinks
>>>   needed. This is same behavior of 3.0 or before.
>>>
>>>   NEVER_DEGENRATE: never degerate this node.
>>>
>>>   DEGENERATE_IF_NOT_PROMOTED: in the master/slave mode, degenrate the
>>>   if it's not chosen as NEVER_DEGENRATE: never degerate this nodethe promoting node.
>>>   (I'm not sure this should apply to slony mode)
>> Not sure that we really need all of that. In my opinion, we binary use
>> or not use autorecovery of slaves. What's a reason where we could decide
>> to not recover a particular slave ?
> There are at least two use cases:
>
> 1) For some reasons (for example, protected by other HA solution such
>    as heartbeat) we do not want to degenerate the node(probably it's
>    the master/primary node). We could have yet another switch
>    "never_degenerate_master_node" or some such, but IMO my proposal is
>    more general and covers the case.

In heartbeat solution, the failed node may be dead and the new master
will not be degenerated as it will be the new master. In all case the
follow master only affect the standby nodes.

Maybe we can have an option to not try recovering the failed node and as
I probably don't saw all the case an option by node allowing or not the
follow the master can bee added too.

> 2) Mostly read only huge database which has multiple standbys. The
>    database is so huge, we don't wan't to recover non promoted
>    standbys until the maintenance time.
This is the default case, i-e : no follow.


Ok, it will be added to the patch. Thanks for your comments.

Regards,

-- 
Gilles Darold
http://dalibo.com - http://dalibo.org