On Tue, Apr 08, 2014 at 12:46:37AM +0200, Stefan Keller wrote:
> Hi Andrew
>
> 2014-04-07 23:37 GMT+02:00 Andrew Sullivan wrote:
> > (1) this has been discussed many times in the past (...)
>
> Can you point me to one of these discussions?
>
> Actually, I browsed once again the mailing list and this is one of the few
> posts I found:
> "In-Memory Columnar Store" 9.12.13 by knizhnik.
I think you have to go back in time further than that. I recall
in-memory table pinning being a recurrent topic during the 8.x series.
I also seem to recall it being mostly on the hackers list.
> I said, that I'd like to discuss things before I code.
Ok. I think in the past what has been successful is some basic design
combined with a POC or some such, generally discussed on -hackers
since that's where all the people who really know the back end hang
out. It seems to me that most of the "built in replication" stuff
that ended up happening worked that way, and that seems to be roughly
similar size of work to this, but I haven't thought about it too much.
> To me it's unclear why design of Postgres should prevent implementation of
> "in-memory tables" e.g. as foreign data wrappers (see e.g. white papers
> for SQL Server mentioned before).
I don't think it does. But new code in the back end isn't free: it
presents a future maintenance burden that others may not be willing to
pay. These things always have to be traded off.
Best regards,
A
--
Andrew Sullivan
ajs@crankycanuck.ca