On Mon, Mar 3, 2014 at 4:12 PM, Robert Haas <robertmhaas@gmail.com> wrote:
> Unfortunately, I don't believe that it's possible to do this easily
> today because of the way bucket splits are handled. I wrote about
> this previously here, with an idea for solving the problem:
We could just tackle this in the same incomplete, buggy, way that
btrees tackled it for years until Heikki fixed them and the way gin
and gist still do I believe. Namely just emit xlog records for each
page individually and during replay remember when you have an
"incomplete split" and complain if recovery ends with any still
incomplete. That would be unfortunate to be adding new cases of this
just as Heikki and company are making progress eliminating the ones we
already had but that's surely better than having no recovery at all.
--
greg