On Tue, May 13, 2003 at 04:55:01PM +0200, Reinoud van Leeuwen wrote:
> One thing comes in mind: portability. Scripts and cron work different on
> Unix compared to Windows or other platforms. Even cron is not the same on
> al Unix variants.
> When the scheduling system is inside the database, it works identical on
> all platforms...
Unless, of course, they have used different versions of an
identically-named library. In which case you get different
performance anyway, and so then you end up writing a custom library
which is or is not the default for HP-UX version 9 when compiled with
certain options (see another recent thread about exactly such a
case).
That's exactly the sort of terrible maintenance problem that I can
see by implementing such functionality, and I can't see that it's
anywhere near worth the cost. Given that the behaviour of /bin/ksh
and cron are both POSIX, you can still rely on some standardisation
across platforms.
It seems to me that, if the price of supporting Windows is that
Postgres has to have its own cron, the cost is too high. I don't
believe that Postgres _does_ need that, however: a scheduling service
is available on Windows that's good enough for these purposes, and
you cannot really expect perfect portability between any flavour of
UNIX and Windows (as anyone who's had to support such a heterogenous
environment knows).
Of course it's true that if you re-implement every service of every
supported operating system yourself, you get a more portable system.
But in that case, perhaps someone should start the PostgrOS project.
(It's a database! No, it's an operating system! No, it's a
data-based operating environment! Wait. Someone already did that:
PICK. Nice system, but not SQL.)
A
--
----
Andrew Sullivan 204-4141 Yonge Street
Liberty RMS Toronto, Ontario Canada
<andrew@libertyrms.info> M2P 2A8 +1 416 646 3304
x110