Horton

Re: Adding skip scan (including MDAM style range skip scan) to nbtree

Peter Geoghegan <pg@bowt.ie>

View thread
From: Peter Geoghegan <pg@bowt.ie>
To: Tomas Vondra <tomas@vondra.me>
Cc: Masahiro.Ikeda@nttdata.com, pgsql-hackers@lists.postgresql.org, Masao.Fujii@nttdata.com
Date: 2024-09-20T14:21:29Z
Lists: pgsql-hackers

Commits

Same data as JSON: GET /api/v1/messages/:b64id/commits the thread's linked commits as JSON, with link sources. API reference →

  1. nbtree: Always set skipScan flag on rescan.

  2. meson: Build numeric.c with -ftree-vectorize.

  3. Fix "variable not found in subplan target lists" in semijoin de-duplication.

  4. Revert "nbtree: Remove useless row compare arg."

  5. nbtree: Remove useless row compare arg.

  6. Prevent premature nbtree array advancement.

  7. nbtree: tighten up array recheck rules.

  8. Avoid treating nonrequired nbtree keys as required.

  9. Adjust overstrong nbtree skip array assertion.

  10. Make NULL tuple values always advance skip arrays.

  11. Avoid extra index searches through preprocessing.

  12. Improve nbtree skip scan primitive scan scheduling.

  13. Further optimize nbtree search scan key comparisons.

  14. Add nbtree skip scan optimization.

  15. Improve nbtree array primitive scan scheduling.

  16. nbtree: Make BTMaxItemSize into object-like macro.

  17. Show index search count in EXPLAIN ANALYZE, take 2.

  18. Make parallel nbtree index scans use an LWLock.

  19. Show index search count in EXPLAIN ANALYZE.

  20. Avoid nbtree parallel scan currPos confusion.

  21. nbtree: Remove useless 'strat' local variable.

  22. Normalize nbtree truncated high key array behavior.

  23. Refactor handling of nbtree array redundancies.

  24. Fix nbtree pgstats accounting with parallel scans.

  25. Avoid parallel nbtree index scan hangs with SAOPs.

  26. Show Parallel Bitmap Heap Scan worker stats in EXPLAIN ANALYZE

  27. Enhance nbtree ScalarArrayOp execution.

  28. Skip checking of scan keys required for directional scan in B-tree

  29. Instead of using a numberOfRequiredKeys count to distinguish required

On Fri, Sep 20, 2024 at 9:45 AM Tomas Vondra <tomas@vondra.me> wrote:
> 3) restart cluster, drop caches
>
> 4) run the query from the SQL script
>
> I suspect you don't do (3). I didn't mention this explicitly, my message
> only said "with uncached data", so maybe that's the problem?

You're right that I didn't do step 3 here. I'm generally in the habit
of using fully cached data when testing this kind of work.

The only explanation I can think of is that (at least on your
hardware) OS readahead helps the master branch more than skipping
helps the patch. That's surprising, but I guess it's possible here
because skip scan only needs to access about every third page. And
because this particular index was generated by CREATE INDEX, and so
happens to have a strong correlation between key space order and
physical block order. And probably because this is an index-only scan.

> I wasn't suggesting it's a sympathetic case for skipscan. My point is
> that it perfectly matches the costing assumptions, i.e. columns are
> independent etc. But if it's not sympathetic, maybe the cost shouldn't
> be 1/5 of cost from master?

The costing is pretty accurate if we assume cached data, though --
which is what the planner will actually assume. In any case, is that
really the only problem you see here? That the costing might be
inaccurate because it fails to account for some underlying effect,
such as the influence of OS readhead?

Let's assume for a moment that the regression is indeed due to
readahead effects, and that we deem it to be unacceptable. What can be
done about it? I have a really hard time thinking of a fix, since by
most conventional measures skip scan is indeed much faster here.

-- 
Peter Geoghegan