Re: pg_upgrade failure on Windows Server

On Thu, Mar 19, 2015 at 7:04 AM, Alvaro Herrera
<alvherre(at)2ndquadrant(dot)com> wrote:
> I haven't followed this effort at all. Are we on the same page about
> backpatching, i.e. we want a minimal patch to apply to the back branches
> (which would be 9.0 to 9.4 at this point, I think, right?); and a larger
> patch that would apply to 9.5.

Yes, that's the plan. What we have been arguing here is the refactoring patch.

> Now that I think about this, perhaps it'd be better to apply the same
> patch to all branches, and once that's settled (i.e. buildfarmed until
> we're sure it works) we refactor things in HEAD.

As a back-branch patch, it is possible to simply use the first version
that Asif has added upthread: it duplicates the restricted token stuff
directly in pg_upgrade and pg_resetxlog.

> Speaking of buildfarming the patch -- how are we to actually test this?
> As far as I understand, this code only matters when the server is being
> run by an administrator account, which the buildfarm code doesn't do ...

Well, I don't know. I am sure I could at least involve some internal
QE folks to get automated tests for this stuff, but the tests we do
are tightly integrated with our build infrastructure so this would
just stay internal, still I would get immediately reports if this
breaks and I could report and provide fixes easily. Having a solution
nicely integrated in the buildfarm scripts is something else though, a
build run being done only as one dedicated user.

> So we're going to have to trust testers running stuff manually, right?

I guess so.. Sorry I have no real solution to that.

Regards,
--
Michael