planner: update the correlation adjustment rule of Limit/TopN for TableScan (#26445)

[ti-srebot]

cherry-pick #26445 to release-5.0
You can switch your code base to this Pull Request by using git-extras:

# In tidb repo:
git pr https://github.com/pingcap/tidb/pull/26653

After apply modifications, you can push your change to this PR via:

git push [email protected]:ti-srebot/tidb.git pr/26653:release-5.0-51c48d2fa287

---

What problem does this PR solve?

Issue Number: close #26088

Problem Summary: planner: update the correlation adjustment rule about Limit/TopN for TableScan

What is changed and how it works?

For TableScan with Limit, the original estimation formula is LimitNum / Selectivity.
For select * from t where year=2003 limit 1, its est-row is 1/est(year=2003).
The formula is based on uniform assumption that we can get one row meeting the condition year=2003 after scanning 1/est(year=2003) rows.

This assumption might be broken when the column year has a high correlation with the PK.
For example, this table has 8 rows: (PK, year): (1, 2002), (2, 2002), (3, 2002), (4, 2002), (5, 2003), (6, 2003), (7, 2003), (8, 2003).
The est-row calculated by the formula is 1/est(year=2003) = 1/0.5 = 2, but actually the act-row is 4 + 1 = 5, since we have to scan through 4 rows with year=2002 before accessing a row with year=2003.

To mitigate this problem, a correlation adjustment rule was introduced before.
In the case above, the est-row will be set to count(year<2003) + 1.

But this rule also brings some risks that it may under-estimate TableScan's row count.
For example, if rows in the table are (PK, year): (0, 2000), (1, 2001), (2, 2002), (3, 2003), (4, 2004), (5, 2005), (6, 2006), (7, 1999).
In this table, year is not strictly ordered, but it still has a high correlation with PK, and the correlation value is larger than our threshold.
Then the est-row of where year=1999 limit 1 will be set to count(year<1999) + 1 which is 1 and under-estimated extremely.

This under-estimation may mislead the optimizer to use TableScan instead of IndexScan(year).
Multiple issues like this case are found, so for safety, we decided to only allow this rule to adjust the upper bound of TableScan, in other words, the est-row is max(LimitNum/Selectivity, AdjustedCount).

Check List

Tests

• Unit test

Release note

• planner: update the correlation adjustment rule about Limit/TopN for TableScan

[ti-chi-bot]

[REVIEW NOTIFICATION]

This pull request has not been approved.

To complete the pull request process, please ask the reviewers in the list to review by filling /cc @reviewer in the comment.
After your PR has acquired the required number of LGTMs, you can assign this pull request to the committer in the list by filling /assign @committer in the comment to help you merge this pull request.

The full list of commands accepted by this bot can be found here.

Reviewer can indicate their review by submitting an approval review.
Reviewer can cancel approval by submitting a request changes review.

[ti-chi-bot]

@ti-srebot: Adding the "do-not-merge/release-note-label-needed" label because no release-note block was detected, please follow our release note process to remove it.

Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes/test-infra repository.

[ti-chi-bot]

@ti-srebot: This cherry pick PR is for a release branch and has not yet been approved by release team.
Adding the do-not-merge/cherry-pick-not-approved label.

To merge this cherry pick, it must first be approved by the collaborators.

AFTER it has been approved by collaborators, please ping the release team in a comment to request a cherry pick review.

Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes/test-infra repository.

[ti-srebot]

/run-all-tests

[ti-srebot]

@qw4990 you're already a collaborator in bot's repo.

[qw4990]

After discussion, we decided not to pick it.