Tom,
> My questions about whether to adopt it have more to do with
> cost/benefit. I haven't seen the patch, but it sounds like it will be
> large and messy; and it's for a feature that nobody ever heard of before,
> let alone one that the community has developed a consensus it wants.
> I'm not interested in adopting stuff just "because DB2 hasn't got it".
OK, to make it a clearer case: we have an increasing user base using
PostgreSQL for decision support. One of the reasons for this is that PG is
the *only* OSDB which does a decent job of DSS. Adding unique DSS features
will make PostgreSQL attractive to a lot more DSS application developers, and
help make up for the things which we don't have yet (parallel query, async
I/O, windowing functions).
"Approximate queries" is something with DSS users *want*. Jim Grey addressed
this in his ACM editiorial on the databases of the future. It's something
that *I* want, and if the Greenplum people aren't speaking up here, it's
because they're not paying atttention.
Now, I don't know if this Skyline patch is our answer for approximate queries.
Maybe I should pester Meredith about getting QBE free of its IP issues; it
certainly looked more flexible than Skyline. In either case, the code
probably needs a complete refactor.
But I think that approximate queries ought to be on our TODO list.
--
Josh Berkus
PostgreSQL @ Sun
San Francisco
