Re: WIP: Covering + unique indexes.

Teodor Sigaev <teodor@sigaev.ru>

From: Teodor Sigaev <teodor@sigaev.ru>
To: Peter Geoghegan <pg@bowt.ie>
Cc: Jeff Janes <jeff.janes@gmail.com>, Alexander Korotkov <a.korotkov@postgrespro.ru>, Anastasia Lubennikova <a.lubennikova@postgrespro.ru>, PostgreSQL Hackers <pgsql-hackers@lists.postgresql.org>
Date: 2018-04-12T16:21:42Z
Lists: pgsql-hackers

Peter Geoghegan wrote:
> On Tue, Apr 10, 2018 at 5:45 PM, Peter Geoghegan <pg@bowt.ie> wrote:
>> I did find another problem, though. Looks like the idea to explicitly
>> represent the number of attributes directly has paid off already:
>>
>> pg@~[3711]=# create table covering_bug (f1 int, f2 int, f3 text);
>> create unique index cov_idx on covering_bug (f1) include(f2);
>> insert into covering_bug select i, i * random() * 1000, i * random() *
>> 100000 from generate_series(0,100000) i;
>> DEBUG:  building index "pg_toast_16451_index" on table "pg_toast_16451" serially
>> CREATE TABLE
>> DEBUG:  building index "cov_idx" on table "covering_bug" serially
>> CREATE INDEX
>> ERROR:  tuple has wrong number of attributes in index "cov_idx"
> 
> Actually, this was an error on my part (though I'd still maintain that
> the check paid off here!). I'll still add defensive assertions inside
> _bt_newroot(), and anywhere else that they're needed. There is no
> reason to not add defensive assertions in all code that handles page
> splits, and needs to fetch a highkey from some other page. We missed a
> few of those.
Agree, I prefer to add more Assert, even. may be, more than actually 
needed. Assert-documented code :)

> 
> I'll add an item to "Decisions to Recheck Mid-Beta" section of the
> open items page for this patch. We should review the decision to make
> a call to _bt_check_natts() within _bt_compare(). It might work just
> as well as an assertion, and it would be unfortunate if workloads that
> don't use covering indexes had to pay a price for the
> _bt_check_natts() call, even if it was a small price. I've seen
> _bt_compare() appear prominently in profiles quite a few times.
> 

Could you show a patch?

I think, we need move _bt_check_natts() and its call under 
USE_ASSERT_CHECKING to prevent performance degradation. Users shouldn't 
pay for unused feature.
-- 
Teodor Sigaev                      E-mail: teodor@sigaev.ru
                                       WWW: http://www.sigaev.ru/


Commits

  1. Adjust INCLUDE index truncation comments and code.

  2. Add commentary explaining why MaxIndexTuplesPerPage calculation is safe.

  3. Indexes with INCLUDE columns and their support in B-tree

  4. Add amcheck verification of heap relations belonging to btree indexes.

  5. Doc: move info for btree opclass implementors into main documentation.

  6. Doc: mention that you can't PREPARE TRANSACTION after NOTIFY.

  7. Remove dedicated B-tree root-split record types.

  8. Restructure index access method API to hide most of it at the C level.

  9. Split _bt_insertonpg to two functions.

  10. Major overhaul of btree index code. Eliminate special BTP_CHAIN logic for