Claudio Natoli wrote:
>
> I wrote:
> > But I'll happily concede the point, and prove it by knocking
> > up a patch for it over the weekend (unless anyone else
> > particularly wants to).
>
> Occurs to me I could kill 2 birds with one stone, and also implement another
> Win32 sticking point, namely the waitpid changes for the Postmaster, by
> having win32_forkexec do one of the following:
>
> a) - when a backend startup is indicated, add a pid/cancel_key struct
> (Backend) to this new array in shared mem
> - when any child of the postmaster is started, also add a pid/HANDLE
> struct to a postmaster local array (or perhaps a dlllist)
>
> b) - when any child of the postmaster is started, add a
> pid/cancel_key/HANDLE/isBackend struct to this new array in shared mem
>
> (HANDLE for waiting on to determine child death; isBackend to separate
> BackendList backends from other children)
>
> Choosing a over b:
> PRO: as we've already been through, keeps the postmaster-only data
> local to the postmaster, stopping tampering from rouge backends
> CON: yet more redundancy
>
> Given recent conversations, I'm guessing (a), but any comments before going
> ahead and doing it?
As I understand it, the postmaster shared memory idea is good because
only the postmaster writes to it, and only the backends read from it.
If the HANDLE works the same way, I think you should put it into the
shared memory too, hence (b).
[ FYI, I haven't seen you on IM lately.]
--
Bruce Momjian | http://candle.pha.pa.us
pgman@candle.pha.pa.us | (610) 359-1001
+ If your life is a hard drive, | 13 Roberts Road
+ Christ can be your backup. | Newtown Square, Pennsylvania 19073
