On Wed, Jul 27, 2016 at 02:33:54PM -0500, Kevin Grittner wrote:
> Until you get to the end of the upgrade and *start the cluster
> under the new version* you can fall back to the old version.
Yeah, but to be fair a lot of well-funded businesses (note what
started this discussion) are pico-managed by people way up the stack
who want a bite-sized answer. If you do not have a happy story for,
"What if we're 48 hours into the upgrade and discover some critical
bug corner case that makes us need to roll back?" then you're going to
lose them. Never mind that such cases literally never happen (if you
have a 48 hour old bug in an Internet system today, you have an
emergency bugfix, not a rollback).
A great deal of practical delivery of technology involves managing
expectations of management who do not understand what they are asking
for and basically want a glib happy answer. As people delivering such
things, we must find a glib happy answer that does not get us fired if
it turns out to be false. The poor story Postgres has about
downgrade, _even if it's a stupid problem_, is a problem. It might
not be worth fixing because it's a stupid problem. But one has to
face the critique in its own terms.
A
--
Andrew Sullivan
ajs@crankycanuck.ca