> I'm adding another table to my replication cluster, and created a new
> set with a single table in it.  The initial data copy failed because
> of some other transaction still being open, so it delayed.  The
> curious part is when the second (and every subsequent0 copy also
> failed, it issued a warning about "no transactin in progress" as shown
> below.  That warning is issued on every subsequent failure to
> subscribe the set.  (Turns out to be a vacuum that started just a
> minute prior to my subscribe!)

> 2007-12-30 12:03:03.743469500 DEBUG1 copy_set 3
> 2007-12-30 12:03:03.752889500 DEBUG1 remoteWorkerThread_4: connected
> to provider DB
> 2007-12-30 12:03:03.759128500 WARN   remoteWorkerThread_4:
> transactions earlier than XID 553658082 are still in progress
> 2007-12-30 12:03:03.759269500 WARN   remoteWorkerThread_4: data copy
> for set 3 failed - sleep 60 seconds
> 2007-12-30 12:03:03.759571500 WARNING:  there is no transaction in
> progress

That WARNING message is consistent with running ROLLBACK against the
provider node after not having done *any* work.

It's probably plausible to hunt this down, and if there has been no work,
to not bother with the ROLLBACK, however, I don't think it fixes terribly
much.

But from a "paranoid programming" perspective, I'd rather be a little
paranoid and *always* ROLLBACK at the end of the COPY_SET attempt, even at
the cost of the (not worrisome) warnings that you're seeing here.