Re: fork/exec patch

Tom Lane wrote:
> Bruce Momjian <pgman(at)candle(dot)pha(dot)pa(dot)us> writes:
> > Agreed, added to the Win32 status page:
> > * remove per-backend parameter file and move into shared memory
>
> [itch] I'm not sure that's an answer either; see my comments about how
> the postmaster shouldn't depend on the contents of shared memory being
> valid.
>
> We could get away with the postmaster having a write-only relationship
> to shared memory (put value of variable X into predetermined location
> Y), but I don't think that helps. It doesn't work for variable-size
> values --- we certainly don't want the postmaster dependent on memory
> allocation structures being valid within shared memory --- and what
> about locks? Do you want the postmaster writing shared values without
> taking a lock, or relying on shared-memory lock structures to be valid
> enough to not lock it up or crash it? My answer to either of those is
> "no way, Jose" ...
>
> Writing temp files may actually be a cleaner solution than writing
> shared memory, once we take these considerations into account. My gripe
> about race conditions was "I want to see how you solve this", and wasn't
> intended to mean "I don't think that is soluble".

Read my idea that shared memory for signals might be required, and a
separate shared memory segment might be used for parameter passing too.

I added a question mark to the win32 TODO item, so we can keep as an
open item.

--
Bruce Momjian | http://candle.pha.pa.us
pgman(at)candle(dot)pha(dot)pa(dot)us | (610) 359-1001
+ If your life is a hard drive, | 13 Roberts Road
+ Christ can be your backup. | Newtown Square, Pennsylvania 19073