Re: Clog handling on exec'ed backend

Tom Lane wrote:
> Bruce Momjian <pgman(at)candle(dot)pha(dot)pa(dot)us> writes:
> > It basically makes one of the Clog data structures into a shared memory
> > area when using exec (EXEC_BACKEND is defined). It passes all the
> > regression/initdb tests on Unix.
>
> It would be cleaner to make the array of LWLockIds part of ClogCtlData,
> if you're going to do it that way. I have doubts that pushing all
> LWLockIds into shared memory is a feasible solution, however.

OK, we will keep it as proposed so that shared memory issue only effects
exec() setups.

> Out of curiosity: have you actually tested any of this Windows code
> you're proposing to commit? A lot of it looks to me like it's not
> gonna work --- stuff is getting done too soon or too late. For
> example, I don't see how a backend is going to read GUC variables
> when it doesn't yet know where the data directory is.

I have not tested it yet because it _all_ has to be done before anything
is going to work. That's why I am trying exec() on Unix first as at
least a minimal way to test things. I am going to have to move things
around once I have the code working.

--
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