Horton

Re: [PATCH] Incremental sort (was: PoC: Partial sort)

Tomas Vondra <tomas.vondra@2ndquadrant.com>

View thread View on postgr.es
From: Tomas Vondra <tomas.vondra@2ndquadrant.com>
To: James Coleman <jtc331@gmail.com>
Cc: Alvaro Herrera <alvherre@2ndquadrant.com>, Tom Lane <tgl@sss.pgh.pa.us>, Michael Paquier <michael@paquier.xyz>, Rafia Sabih <rafia.pghackers@gmail.com>, Peter Geoghegan <pg@bowt.ie>, Simon Riggs <simon@2ndquadrant.com>, Shaun Thomas <shaun.thomas@2ndquadrant.com>, Dmitry Dolgov <9erthalion6@gmail.com>, Alexander Korotkov <a.korotkov@postgrespro.ru>, PostgreSQL Developers <pgsql-hackers@lists.postgresql.org>, Andreas Karlsson <andreas@proxel.se>
Date: 2020-04-06T22:13:54Z
Lists: pgsql-hackers 
On Mon, Apr 06, 2020 at 05:47:48PM -0400, James Coleman wrote:
>On Mon, Apr 6, 2020 at 5:40 PM Tomas Vondra
><tomas.vondra@2ndquadrant.com> wrote:
>>
>> On Mon, Apr 06, 2020 at 11:12:32PM +0200, Tomas Vondra wrote:
>> >On Mon, Apr 06, 2020 at 04:54:38PM -0400, Alvaro Herrera wrote:
>> >>On 2020-Apr-06, Tom Lane wrote:
>> >>
>> >>>Locally, things pass without force_parallel_mode, but turning it on
>> >>>produces failures that look similar to rhinoceros's (didn't examine
>> >>>other BF members).
>> >>
>> >>FWIW I looked at the eight failures there were about fifteen minutes ago
>> >>and they were all identical.  I can confirm that, in my laptop, the
>> >>tests work without that GUC, and fail in exactly that way with it.
>> >>
>> >
>> >Yes, there's a thinko in show_incremental_sort_info() and it returns too
>> >soon. I'll push a fix in a minute.
>> >
>>
>> OK, I've pushed a fix - this should make the buildfarm happy again.
>>
>> It however seems to me a bit more needs to be done. The fix makes
>> show_incremental_sort_info closer to show_sort_info, but not entirely
>> because IncrementalSortState does not have sort_Done flag so it still
>> depends on (fullsortGroupInfo->groupCount > 0). I haven't noticed that
>> before, but not having that flag seems a bit weird to me.
>>
>> It also seems possibly incorrect - we may end up with
>>
>>    fullsortGroupInfo->groupCount == 0
>>    prefixsortGroupInfo->groupCount > 0
>>
>> but we won't print anything.
>
>This shouldn't ever be possible, because the only way we get any
>prefix groups at all is if we've already sorted a full sort group
>during the mode transition.
>
>> James, any opinion on this? I'd say we should restore the sort_Done flag
>> and make it work as in plain Sort. Or some comment explaining why
>> depending on the counts is OK (assuming it is).
>
>There's previous email traffic on this thread about that (I can look
>it up later this evening), but the short of it is that I believe that
>relying on the group count is actually more correct than a sort_Done
>flag in the case of incremental sort (in contrast to regular sort).
>

OK. Maybe we should add a comment to explain.c saying it's OK.

I've pushed a fix for failures due to different planned workers (in the
test I added to show changes due to add_partial_path tweaks).

It seems we're not out of the woods yet, though. rhinoceros and
sidewinder failed with something like this:

                 Sort Method: quicksort  Memory: NNkB
+               Sort Method: unknown  Disk: NNkB

Would you mind investigating at it?

regards

-- 
Tomas Vondra                  http://www.2ndQuadrant.com
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services

Commits

  1. Further adjustments to Hashagg EXPLAIN ANALYZE output

  2. Rework EXPLAIN format for incremental sort

  3. Fix typos and improve incremental sort comments

  4. Stabilize incremental_sort tests

  5. Minor improvements in Incremental Sort explain

  6. Consider Incremental Sort paths at additional places

  7. Fix representation of SORT_TYPE_STILL_IN_PROGRESS.

  8. Fix failures in incremental_sort due to number of workers

  9. Fix show_incremental_sort_info with force_parallel_mode

  10. Implement Incremental Sort

  11. Fix handling of "Subplans Removed" field in EXPLAIN output.

  12. Fix EXPLAIN (SETTINGS) to follow policy about when to print empty fields.

  13. Ensure plpgsql result tuples have the right composite type marking.

  14. Propagate sort instrumentation from workers back to leader.

  15. Make new regression test case parallel-safe, and improve its output.

  16. Push limit through subqueries to underlying sort, where possible.

  17. Fix inappropriate printing of never-measured times in EXPLAIN.

  18. Fix some infelicities in EXPLAIN output for parallel query plans.