"Troy Wolf" <troy at troywolf.com> writes:
>           I'm looking for suggestions as to the best way to verify that my single
>      slave is in sync with my master.  We've had a couple of situations (self
>      induced) where we've created problems for slony and therefore would like
>      to provide a warm and fuzzy feeling that slony is current.
>      
>
> I thought I was clever and built a script that selects the most recent timestamp from the event tables on both and compares to see that they are
> within 10 seconds. It took me about 45 days to learn the hard way that this is not enough.
>
> So I got stupid-brute-force. I created a cron job that updates a value in a slony'd table's column every minute to the current_timestamp. My
> "health" check simply queries the subscriber's value for this column to see if it is within the last 2 minutes.  For me, I had an appropriate
> table of "parameters" where I could just create a "slony_ts" parameter and update its value. You may not have this, so maybe you'd create a
> special table just for this purpose.

There is value to both approaches:

 - Checking sl_status to see if Slony-I thinks it's behind will work
   fairly well both when things are working well, and when your
   underlying system *isn't* busy (e.g. - when it's not doing any
   "real work."

 - There is also value to having an "end-to-end" test that looks for
   some application data.

   We have a "transaction table" which is normally updated quite
   frequently, 24 hours a day.  Looking at the youngest transaction in
   that table is a pretty good test; if there's no new-looking data,
   then there's probably a problem with the *application*, even if
   replication's working fine.
-- 
(reverse (concatenate 'string "ofni.secnanifxunil" "@" "enworbbc"))
http://linuxfinances.info/info/spiritual.html
The  meta-Turing test counts  a thing  as intelligent  if it  seeks to
apply Turing tests to objects of its own creation.