Ah, sorry for the late response (and for any confusion), but the only
thing I tested was Qingqing's rewritten semaphore implementation. I
didn't test the proposed bug fixes to the existing semaphore
implementation.
We've never been able reproduce (or even trigger) the original "sem_ctl
failed" error in a testing environment so it would be hard to say if the
changes to win32/sema.c have an impact on it or not. On the other hand,
win32_sema.c seems to solve the pgbench lockups reported earlier by Jim
N. and it successfully completes a reasonably brutal stress test with
real world data and real world queries (which at least is a good
indication that it basically works).
Pete
>>> Bruce Momjian <pgman@candle.pha.pa.us> 05/11/06 4:08 am >>>
Tom Lane wrote:
> Bruce Momjian <pgman@candle.pha.pa.us> writes:
> >> Well we did have a bug report by Peter Brant, and a test by him
with the
> >> patch that fixes it, so it seems it should be applied. The URLs
I
> >> posted had that information.