On 04/10/2017 12:25 PM, Michael Banck wrote:
> Am Montag, den 10.04.2017, 11:42 -0400 schrieb Peter Eisentraut:
>> On 4/8/17 04:50, Michael Banck wrote:
>>> "Client-Side Connection-Failover"
>>
>> I don't think the libpq feature can be considered "failover".
>
> Hrm.
It's still valuable, though, especially in combination with replication
which offers rapid failover. It means that instead of needing to update
all of your clients with a new list of servers immediately, you can wait
and assume that their reconnection logic will handle things until an
admin can push out an update.
Particularly, with MM replication (like BDR or MMSync), it means you
don't have to worry about failover on the clients at all.
So it's not a major feature on its own, but it's a good part of a
scale-out story.
--
Josh Berkus
Containers & Databases Oh My!