Hi,
On 2020-06-23 21:50:26 -0400, Chapman Flack wrote:
> On 06/23/20 21:44, Andres Freund wrote:
>
> > I think that's way harder than what you make it sound here. The locking
> > for shm_mq doesn't really work inside a process. In contrast to the
> > single threaded case something like a volatile write to
> > ParallelMessagePending doesn't guarantee much, because there's no
> > guaranteed memory ordering between threads. And more.
>
> It occurred to me after I sent the message this morning that my suggestion
> (2) could subsume (1). And requires nothing more than a single volatile
> write of a boolean, and getting called back at a convenient time on the
> single main thread.
A single volatile write wouldn't guarantee you much in the presence of
multiple threads. You could very well end up with a concurrent
CHECK_FOR_INTERRUPTS() in the main thread unsetting InterruptPending,
but not yet seeing / processing ParallelMessagePending. Nor would it
wake up the main process if it's currently waiting on a latch.
Greetings,
Andres Freund