Re: Isn't wait_for_catchup slightly broken?

I wrote:
> Another thing that is bothering me a bit is that a number of the
> callers use $node->lsn('insert') as the target.  This also seems
> rather dubious, because that could be ahead of what's been written
> out.  These callers are just taking it on faith that something will
> eventually cause that extra WAL to get written out (and become
> available to the standby).  Again, that seems to make the test
> slower than it need be, with a worst-case scenario being that it
> eventually times out.  Admittedly this is unlikely to be a big
> problem unless some background op issues an abortive transaction
> at just the wrong time.  Nonetheless, I wonder if we shouldn't
> standardize on "thou shalt use the write position", because I
> don't think the other alternatives have anything to recommend them.

Here's a version that makes sure that callers specify a write position not
an insert position.  I also simplified the callers wherever it turned
out that they could just use the default parameters.

			regards, tom lane