Jerry,
* Jerry Sievers (gsievers19@comcast.net) wrote:
> Bruce Momjian <bruce@momjian.us> writes:
> > I agree, but I am not sure how to improve it. The big complaint I have
> > heard is that once you upgrade and open up writes on the upgraded
> > server, you can't re-apply those writes to the old server if you need to
> > fall back to the old server. I also don't see how to improve that either.
>
> Hmmm, is it at least theoretically possible that if a newly upgraded
> system were run for an interval where *no* incompatible changes to DDL
> etc had been done...
>
> ...that a downgrade could be performed?
>
> Er, using a not yet invented pg_downgrade:-)
The short answer is 'no'. Consider a case like the GIN page changes- as
soon as you execute DML on a column that has a GIN index on it, we're
going to rewrite that page using a newer version of the page format and
an older version of PG isn't going to understand it.
Those kind of on-disk changes are, I suspect, why you have to set the
"compatibility" option in the big $O product to be able to later do a
downgrade.
> That is, since higher version knew enough about lower version to
> rejigger everything... just maybe it could do the reverse.
That might work if you opened the database in read-only mode, but not
once you start making changes.
Thanks!
Stephen