Horton

Re: remaining sql/json patches

amit <amitlangote09@gmail.com>

View thread
From: Amit Langote <amitlangote09@gmail.com>
To: Tomas Vondra <tomas.vondra@enterprisedb.com>
Cc: Alvaro Herrera <alvherre@alvh.no-ip.org>, Erik Rijkers <er@xs4all.nl>, jian he <jian.universality@gmail.com>, Andres Freund <andres@anarazel.de>, Andrew Dunstan <andrew@dunslane.net>, PostgreSQL-development <pgsql-hackers@postgresql.org>
Date: 2024-03-06T04:07:33Z
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. SQL/JSON: Various improvements to SQL/JSON query function docs

  2. SQL/JSON: Fix some obsolete comments.

  3. SQL/JSON: Fix issues with DEFAULT .. ON ERROR / EMPTY

  4. JSON_TABLE: Add support for NESTED paths and columns

  5. Fix JsonExpr deparsing to emit QUOTES and WRAPPER correctly

  6. Fix typo introduced in 6185c9737

  7. Add basic JSON_TABLE() functionality

  8. Avoid splitting errmsg string to span multiple lines

  9. Add SQL/JSON query functions

  10. Implement various jsonpath methods

  11. Add soft error handling to some expression nodes

  12. Adjust populate_record_field() to handle errors softly

  13. Refactor code used by jsonpath executor to fetch variables

  14. Test EXPLAIN (FORMAT JSON) ... XMLTABLE

  15. Simplify productions for FORMAT JSON [ ENCODING name ]

  16. Add trailing commas to enum definitions

  17. doc: add missing <returnvalue> and whitespace

  18. Add more SQL/JSON constructor functions

  19. Rename a nonterminal used in SQL/JSON grammar

  20. Some refactoring to export json(b) conversion functions

  21. Don't include CaseTestExpr in JsonValueExpr.formatted_expr

  22. Code review for commit b6e1157e7d

  23. Pass constructName to transformJsonValueExpr()

  24. Unify JSON categorize type API and export for external use

  25. Make some indentation in gram.y consistent

  26. Allow most keywords to be used as column labels without requiring AS.

  27. Reduce size of backend scanner's tables.

  28. Use perfect hashing, instead of binary search, for keyword lookup.

Hi Tomas,

On Wed, Mar 6, 2024 at 6:30 AM Tomas Vondra
<tomas.vondra@enterprisedb.com> wrote:
>
> Hi,
>
> I know very little about sql/json and all the json internals, but I
> decided to do some black box testing. I built a large JSONB table
> (single column, ~7GB of data after loading). And then I did a query
> transforming the data into tabular form using JSON_TABLE.
>
> The JSON_TABLE query looks like this:
>
> SELECT jt.* FROM
>   title_jsonb t,
>   json_table(t.info, '$'
>     COLUMNS (
>       "id" text path '$."id"',
>       "type" text path '$."type"',
>       "title" text path '$."title"',
>       "original_title" text path '$."original_title"',
>       "is_adult" text path '$."is_adult"',
>       "start_year" text path '$."start_year"',
>       "end_year" text path '$."end_year"',
>       "minutes" text path '$."minutes"',
>       "genres" text path '$."genres"',
>       "aliases" text path '$."aliases"',
>       "directors" text path '$."directors"',
>       "writers" text path '$."writers"',
>       "ratings" text path '$."ratings"',
>       NESTED PATH '$."aliases"[*]'
>         COLUMNS (
>           "alias_title" text path '$."title"',
>           "alias_region" text path '$."region"'
>         ),
>       NESTED PATH '$."directors"[*]'
>         COLUMNS (
>           "director_name" text path '$."name"',
>           "director_birth_year" text path '$."birth_year"',
>           "director_death_year" text path '$."death_year"'
>         ),
>       NESTED PATH '$."writers"[*]'
>         COLUMNS (
>           "writer_name" text path '$."name"',
>           "writer_birth_year" text path '$."birth_year"',
>           "writer_death_year" text path '$."death_year"'
>         ),
>       NESTED PATH '$."ratings"[*]'
>         COLUMNS (
>           "rating_average" text path '$."average"',
>           "rating_votes" text path '$."votes"'
>         )
>     )
>   ) as jt;
>
> again, not particularly complex. But if I run this, it consumes multiple
> gigabytes of memory, before it gets killed by OOM killer. This happens
> even when ran using
>
>   COPY (...) TO '/dev/null'
>
> so there's nothing sent to the client. I did catch memory context info,
> where it looks like this (complete stats attached):
>
> ------
> TopMemoryContext: 97696 total in 5 blocks; 13056 free (11 chunks);
>                   84640 used
>   ...
>   TopPortalContext: 8192 total in 1 blocks; 7680 free (0 chunks); ...
>     PortalContext: 1024 total in 1 blocks; 560 free (0 chunks); ...
>       ExecutorState: 2541764672 total in 314 blocks; 6528176 free
>                      (1208 chunks); 2535236496 used
>         printtup: 8192 total in 1 blocks; 7952 free (0 chunks); ...
>         ...
> ...
> Grand total: 2544132336 bytes in 528 blocks; 7484504 free
>              (1340 chunks); 2536647832 used
> ------
>
> I'd say 2.5GB in ExecutorState seems a bit excessive ... Seems there's
> some memory management issue? My guess is we're not releasing memory
> allocated while parsing the JSON or building JSON output.
>
> I'm not attaching the data, but I can provide that if needed - it's
> about 600MB compressed. The structure is not particularly complex, it's
> movie info from [1] combined into a JSON document (one per movie).

Thanks for the report.

Yeah, I'd like to see the data to try to drill down into what's piling
up in ExecutorState.  I want to be sure of if the 1st, query functions
patch, is not implicated in this, because I'd like to get that one out
of the way sooner than later.

-- 
Thanks, Amit Langote