Re: INSERT ... ON CONFLICT UPDATE/IGNORE 4.0

On 2015-05-08 15:22:09 -0400, Tom Lane wrote:
> I'm back to suspecting that the indcheckxmin issue is the true cause of
> the buildfarm failure

Me too.

> though we lack an explanation why Andres failed to reproduce it ...

My laptop is probably a good bit faster than jaguarundi, particularly in
a single "threaded" workload, as that test, due to turbo
boost. Friarbird didn't trigger it either so far.  It's quite possible
that it requires a concurrent analyze or so to run to occur in the wrong
moment.

I've pushed the fixes to those two. I guess we'll have to wait a couple
(slow) buildfarm cycles to see whether it fixes things.

Greetings,

Andres Freund