On Sun, Feb 27, 2022 at 11:18 AM Melanie Plageman <melanieplage...@gmail.com> wrote: > How could it be that worker 2 is waiting on the build barrier in > PHJ_BUILD_HASHING_INNER and worker 1 and the leader are waiting on it > with it supposedly in PHJ_BUILD_HASHING_OUTER?
That'd be consistent with a wakeup going missing, somehow. W2 was supposed to be released but somehow didn't get the message (until a random debugger-induced EINTR releases it and it sees all the state it needs to proceed in shared memory), meanwhile the other two got to the end of the next phase and are waiting for it.
