Thanks Jeff for picking this up.
> I plan to commit something close to this patch in about 48 hours and
> backport to 12, like 5f12bc94dc was.
>
Using the repro provided, I can confirm that the patch works on master
(6bcda4a7212) and thus also +1 on going ahead with pushing this.
> I would prefer if I was able to repro on 13 first, and some review
> would be nice given the area involved and how close we are to release.
> But it seems like fixing a bug is better than not.
Outside of the repro steps, I applied this patch on my test-instance
(the rationale was to attempt the same workload that generated the
original set of crashes) and ran the workload but didn't get anything
to report - which isn't conclusive, but IMHO still good.
Given that the original set of crashes were generated over a month of
tracking, I'll keep an eye on the results for the next few weeks and come
back in case I have something interesting.
-
robins