Re: Let's invent a function to report lock-wait-blocking PIDs

Heikki Linnakangas <hlinnakangas@vmware.com> writes:
> On 21.03.2013 05:36, Tom Lane wrote:
>>> The API that comes to mind is (name subject to bikeshedding)
>>> pg_blocking_pids(pid int) returns int[]

> How about inverting the function into:
> pg_pid_blocked_by(pid int) returns int
> It would take as argument a pid, and return the pid of the process that 
> is blocking the given process. That would feel more natural to me.

Hm, I'm not sure that's uniquely defined.  In the case I mentioned
before (A has AccessShare, B is blocked waiting for AccessExclusive,
C wants AccessShare and is queued behind B), which of A and B do
you think is blocking C?

Whichever answer you choose could be the wrong one for isolationtester:
I think it needs to consider that C is blocked if *either* A or B is
part of its set of test processes.  So that's why I thought an array
(or set) result including both A and B would be appropriate.  AFAICT,
what you're proposing isn't the "inverse" of what I said, it's the
same direction but you're assuming there's only one blocking process.
        regards, tom lane