perf(over window): skip remaining affected rows when rank is not changed

[stdrc]

I hereby agree to the terms of the RisingWave Labs, Inc. Contributor License Agreement.

What's changed and what's your intention?

For example:

rank() over (
  partition by ...
  order by val
)

Rows:

 row_id | val | rank
--------|-----|------
 100    | 2   | 1
 101    | 6   | 2
 102    | 7   | 3
 103    | 9   | 4

When changing val of row 101, it will potenrially affect row 102 and 103. Previously, no matter what the val is changed to, we will re-compute all outputs from row 101 to the end of the partition (row 103). After this PR, if the output of row 102 is not changed, we will stop the re-computation.

Checklist

• I have written necessary rustdoc comments
• I have added necessary unit tests and integration tests
• I have added test labels as necessary. See details.
• I have added fuzzing tests or opened an issue to track them. (Optional, recommended for new SQL features Sqlsmith: Sql feature generation #7934).
• My PR contains breaking changes. (If it deprecates some features, please create a tracking issue to remove them in the future).
• All checks passed in ./risedev check (or alias, ./risedev c)
• My PR changes performance-critical code. (Please run macro/micro-benchmarks and show the results.)

• My PR contains critical fixes that are necessary to be merged into the latest release. (Please check out the details)

Documentation

• My PR needs documentation updates. (Please use the Release note section below to summarize the impact on users)

Release note

If this PR includes changes that directly affect users or other significant modifications relevant to the community, kindly draft a release note to provide a concise summary of these changes. Please prioritize highlighting the impact these changes will have on users.

[stdrc]

• perf(over window): skip remaining affected rows when rank is not changed #18950 👈
• perf(over window): pass through Updates not affecting window function outputs #19056
• perf(over window): avoid recompute information about window frames #19036
• fix(over window): fix table iter pk prefix to use deduped partition key #19081
• main

This stack of pull requests is managed by Graphite. Learn more about stacking.

Join @stdrc and the rest of your teammates on Graphite

[stdrc]

After some loose benchmark based on a custom query on nexmark datagen, I found that this PR doesn't help much, but still better than nothing.

Query (no much rationale, just a quick attempt to imitate a user's query):

CREATE VIEW v1 AS
SELECT
    foo,
    bar,
    SUM(point) AS points
FROM (
    SELECT
        ((bidder * 2654435761) # (bidder >> 16)) % 10000 as foo,
        auction % 3 as bar,
        price % 10 as point
    FROM bid
)
GROUP BY foo, bar;

CREATE SINK mv AS
SELECT
    foo,
    bar,
    points,
    ROW_NUMBER() OVER (
        PARTITION BY bar
        ORDER BY points DESC, foo
    ) AS r
FROM v1
WITH ( connector = 'blackhole', type = 'append-only', force_append_only = 'true');

Before:

After:

I'm going to merge this first, since no more operations can be done on this PR and the performance improvement looks small.