Support aggregates and constant filters in `QUALIFY`

[alamb]

@haohuaijin added support for QUALIFY in

• feat: implement QUALIFY clause #16933

@Vedin has pointed out some follow on work here

Hi @haohuaijin, I accidentally worked on the same feature. The approach is the same. So, I don't want anyhow conflict with your contribution and will just wait until your PR is merged. I just want to highlight 2 cases that probably are not currently covered in your PR.

1. Aggregate functions with QUALIFY

CREATE TABLE qt (i INT, p VARCHAR, o INT) AS VALUES
  (1, 'A', 1),
  (2, 'A', 2),
  (3, 'B', 1),
  (4, 'B', 2);

SELECT p, SUM(o) AS s
FROM qt
GROUP BY p
QUALIFY RANK() OVER (ORDER BY s DESC) = 1
ORDER BY p;

2. Constant filter + QUALIFY

CREATE TABLE web_base_events_this_run (
  domain_sessionid VARCHAR,
  app_id VARCHAR,
  page_view_id VARCHAR,
  derived_tstamp TIMESTAMP,
  dvce_created_tstamp TIMESTAMP,
  event_id VARCHAR
) AS SELECT * FROM VALUES
  ('ds1', 'appA', NULL, '2025-01-01 10:00:00'::timestamp, '2025-01-01 10:05:00'::timestamp, 'e1'),
  ('ds1', 'appA', NULL, '2025-01-01 11:00:00'::timestamp, '2025-01-01 11:00:00'::timestamp, 'e2'),
  ('ds1', 'appA', 'pv', '2025-01-01 12:00:00'::timestamp, '2025-01-01 12:00:00'::timestamp, 'e3'),
  ('ds2', 'appB', NULL, '2025-01-01 09:00:00'::timestamp, '2025-01-01 09:10:00'::timestamp, 'e4'),
  ('ds2', 'appB', NULL, '2025-01-01 09:05:00'::timestamp, '2025-01-01 09:09:00'::timestamp, 'e5');
  
  SELECT domain_sessionid, app_id
FROM web_base_events_this_run
WHERE page_view_id IS NULL
QUALIFY ROW_NUMBER() OVER (
  PARTITION BY domain_sessionid
  ORDER BY derived_tstamp, dvce_created_tstamp, event_id
) = 1
ORDER BY domain_sessionid;

I covered the first one by adding qualify expressions to aggr_expr_haystack:

        let aggr_expr_haystack = select_exprs
            .iter()
            .chain(having_expr_opt.iter())
            .chain(qualify_expr_opt_pre_aggr.iter());

The second one required me to change the logic in common_subexpr_eliminate.rs. You can check it out here (Embucket#34). Maybe you'll come up with a better solution.
Hope this helps.

Originally posted by @Vedin in #16933 (comment)

[rkrishn7]

Hello! I've opened #17313 that should fix the aggregate problem.

As for the second concern, the example provided works with the original QUALIFY implementation. Maybe we can add some more context regarding the specific problem?

Thanks!

[Vedin]

Hello! I've opened #17313 that should fix the aggregate problem.

As for the second concern, the example provided works with the original QUALIFY implementation. Maybe we can add some more context regarding the specific problem?

Thanks!

Yeah, the provided example really works for me as well, but there are still some problems with common_sub_expression_eliminate rule. I have tried this example with your changes, and it appears to be failing. So, reproducible is:

CREATE TABLE bulk_import_entities (
  id INT,
  _task_instance INT,
  _uploaded_at TIMESTAMP
) AS VALUES
  (1, 1, '2025-01-01 10:00:00'::timestamp),
  (1, 2, '2025-01-02 09:00:00'::timestamp),
  (1, 2, '2025-01-03 08:00:00'::timestamp),
  (2, 1, '2025-01-01 11:00:00'::timestamp),
  (2, 2, '2025-01-02 12:00:00'::timestamp),
  (3, 1, '2025-01-01 13:00:00'::timestamp);

SELECT id, _task_instance
FROM bulk_import_entities
WHERE _task_instance = (
  SELECT MAX(_task_instance) FROM bulk_import_entities
)
QUALIFY ROW_NUMBER() OVER (PARTITION BY id ORDER BY _uploaded_at) = 1
ORDER BY id;

with error:

Optimizer rule 'common_sub_expression_eliminate' failed
caused by
Error during planning: Window has mismatch between number of expressions (1) and number of fields in schema (0)

While the correct output is

1 2
2 2