Re: UNIQUE capability to hash indexes - Mailing list pgsql-hackers

From Antonin Houska
Subject Re: UNIQUE capability to hash indexes
Date
Msg-id 10392.1454629373@localhost
Whole thread Raw
In response to Re: UNIQUE capability to hash indexes  (Tom Lane <tgl@sss.pgh.pa.us>)
List pgsql-hackers
Tom Lane <tgl@sss.pgh.pa.us> wrote:

> Alvaro Herrera <alvherre@2ndquadrant.com> writes:
> > Tom Lane wrote:
> >> Not that I've heard of.  It's very hard to muster any enthusiasm for
> >> improving hash indexes unless their lack of WAL-logging is fixed first.
>
> > This is really strange though.  Surely adding WAL-logging is not an
> > enormous task anymore ... I mean, we're undertaking far larger efforts
> > now, the WAL logging code is simpler than before, and we even have a
> > tool (ok, gotta streamline that one a little bit) to verify that the
> > results are correct.
>
> ISTR that we discussed this previously and ran into some stumbling block
> or other that made it less-than-trivial.  Don't recall what though.

Concurrency of bucket split is the issue. It makes sense to fix the problem
before anyone tries to implement WAL. Otherwise WAL will have to be reworked
from scratch someday.

I had some ideas which I even published:

http://www.postgresql.org/message-id/32423.1427413442@localhost

Then I spent some more time on it sometime in October and improved the concept
a bit (and also found bugs in the version I had published, so please don't
spend much time looking at it). I also wrote a function to check if the
consistent (to be possibly added to pageinspect extension). I even wrote a
function that inserts tuples only into index, not into heap - I suppose that
should make comparison of index performance with and without the patch
simpler.

Now, besides making the patch easier to read, I need to test it
thoroughly. The lack of time is one problem, but I need to admit that it's a
personal issue too :-) So far I think I have a good idea, but now I should try
hard to break it.

Now that I see the problem mentioned again, I feel myself kind of "ignited".
I expect to have some leisure time at the end of February, so I'll test the
patch and post my results early in March.

--
Antonin Houska
Cybertec Schönig & Schönig GmbH
Gröhrmühlgasse 26
A-2700 Wiener Neustadt
Web: http://www.postgresql-support.de, http://www.cybertec.at



pgsql-hackers by date:

Previous
From: Andreas Karlsson
Date:
Subject: Re: UNIQUE capability to hash indexes
Next
From: Jim Nasby
Date:
Subject: Re: "using previous checkpoint record at" maybe not the greatest idea?