On Mon, May 9, 2011 at 9:29 AM, Frank McGeough
<Frank.Mcgeough at vocalocity.com> wrote:
> Environment : postgres 8.4, slony 1.2.20
>
> When adding a new node to an existing replication environment its my understanding that you would first get the slon process running on the new node. Even without subscribing sets the slon process on the new node would need to be running in order to handle events, correct? Otherwise, the sl_status st_lag_time for this new node would report a larger and larger lag time, right?
>
> I'm asking because someone that I know suggested that you could add a node and subscribe it a couple weeks later and it wouldn't effect replication at all in that interim period. I said that it wouldn't change replication to the existing nodes but that the new node would possibly trigger alarms if they were based on st_lag_time. Even though a new node isn't subscribed it still must handle ongoing events, right? Even if the handling is to discard them? That is, the master sends out events to all slaves and how they process them is decided on the slave slon process - the master doesn't filter things to send different events out to different nodes based on their subscription.

Sure, it'll have further adverse effects, too.

Since that new node has not reported confirmation of processing of
events, the cluster will hold onto *all* replicable data, so that
tables sl_log_1, sl_log_2, and the sequence log will grow unboundedly,
as nothing will be trimming their data.  I think you'll see ballooning
of sl_event and sl_confirm, too, but that won't be nearly as
significant as unbounded growth of sl_log_[12].