Alvaro Herrera <alvherre@commandprompt.com> wrote:
> Excerpts from Kevin Grittner's message of vie ago 31 17:23:20
-0400 2007:
>
>>> I don't know exactly how much it'd take, but I would *love* to
>>> see this as a "proper case study" on our website. I know we
>>> already have a couple of quotes in the archive, but a complete
>>> case study would be even better. WOuld that be doable?
>>
>> I'll poke around and see what other case studies look like. I'm
>> pretty sure management would not approve it as "on the clock"
>> work, but if the effort is within reasonable bounds, I'll do it
>> on my own time.
> Did you ever release some sort of case study for the Wisconsin
> courts database system? Pointers to anything you have would be
> very much appreciated.
No, I never got comfortable with what format to use, but I *did* do
a presentation at PGCon 2009, with slides.
http://www.pgcon.org/2009/schedule/events/129.en.html
The big news since then is that we decided that if we could find a
solution to the serializable versus snapshot isolation issues, we
could commit to PostgreSQL for the long term. This led to the SSI
effort, and with the acceptance of that patch, we moved forward with
converting all of the triggers and many of the "stored procedures"
(1900 in total) which had been running on our Java tier just above
the database into PostgreSQL plpgsql functions. This required some
minor hacking on PostgreSQL core which we are currently "forking",
but has been surprisingly painless. (Again, the "product-neutral"
ANSI SQL format we used, allowed us to parse existing code and walk
the AST to generate PostgreSQL function source code -- but that's
its last gasp; we're maintaining it in plpgsql going forward.)
I submitted some of the hacks from this latest phase to the first CF
for 9.2, and got significant interest and good feedback, but haven't
found the round tuits to follow up on getting these out to the
community with the requested changes. All except the "trigger
depth" patch could and should be done as extensions, now that we
have a mature facility for that. Whether any of those extensions
belong in core is probably going to be a matter of some debate; I
think the "triggered change notification" showed enough interest (on
and off list) that it might belong.
-Kevin
