Re: Logging conflicted queries on deadlocks

Gregory Stark <stark(at)enterprisedb(dot)com> writes:
> Ah, reading the patch I see comments indicating that yes that's possible and
> no, we don't really care. So, ok. If the backend disappears entirely could the
> string be empty?

Right, we'd be pointing at a BackendStatusArray entry that was now
unused, or even perhaps recycled by a new session. That memory doesn't
move, so we don't need to worry about picking up something that's not
a status string at all, but worst case it could be not the string we
want. I think the odds are pretty low though.

Perhaps it would be safer to copy out st_activity inside the
> loop checking st_changecount?

Don't think it would really help any --- the other backend could've
aborted and changed its status string before we ever get to this code at
all.

> It is a really nice feature though. Note that there was an unrelated demand
> for just this info on one of the other lists just today. Thanks very much
> Itagaki-san!

That was my feeling --- the usefulness is high enough that a small
probability of a wrong display is a small price to pay.

regards, tom lane