planner: fix incorrect estRows with global index and json column

[hawkingrei]

What problem does this PR solve?

Issue Number: close #55818

Problem Summary:

What changed and how does it work?

Because the JSON column will be without stats by default, the row count will be 0. But when calculating selectivity, we cannot divide by zero. it will get NaN row count.

why is the result wrong?

It will return NaN from Selectivity and break all stats info by ds.TableStats.Scale(selectivity)

Final, The problem will be created at the joinReorderGreedySolver.solve-> constructConnectedJoinTree

checkConnectionAndMakeJoin will create a right join plan. but when to calculate the cost, it will get NaN because of the pollute stats info. we cannot get best join because of Nan > math.MaxFloat64. makeBushyJoin will not consider the EqCondtion. Many condition will lose. This is the reason why it will get wrong result.

Check List

Tests

• Unit test
• Integration test
• Manual test (add detailed scripts or steps below)

Due to incorrect statistics, the filter condition was completely removed.

before

+---------------------------+---------+-----------+-----------------+------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| id                        | estRows | task      | access object   | operator info                                                                                                                                                                            |
+---------------------------+---------+-----------+-----------------+------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| Projection_12             | NaN     | root      |                 | 1->Column#18                                                                                                                                                                             |
| └─HashJoin_13             | NaN     | root      |                 | CARTESIAN inner join                                                                                                                                                                     |
|   ├─TableReader_19(Build) | 38.00   | root      |                 | data:TableFullScan_18                                                                                                                                                                    |
|   │ └─TableFullScan_18    | 38.00   | cop[tikv] | table:t61a85298 | keep order:false, stats:pseudo                                                                                                                                                           |
|   └─TableReader_17(Probe) | NaN     | root      | partition:all   | data:Selection_16                                                                                                                                                                        |
|     └─Selection_16        | NaN     | cop[tikv] |                 | or(json_memberof(cast(16739493649928310215, json BINARY), test.tceb7972c.col_17), not(istrue_with_null(json_contains(test.tceb7972c.col_17, cast("6019730272580550835", json BINARY))))) |
|       └─TableFullScan_15  | 12.00   | cop[tikv] | table:tceb7972c | keep order:false, stats:partial[col_17:unInitialized]                                                                                                                                    |
+---------------------------+---------+-----------+-----------------+------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+

after

id  estRows task  access object operator info
Projection_11 6.00  root    1->Column#18
└─HashJoin_13 6.00  root    inner join, equal:[eq(test.tceb7972c.col_19, Column#19)]
  ├─TableReader_16(Build) 4.80  root  partition:all data:Selection_15
  │ └─Selection_15  4.80  cop[tikv]   or(json_memberof(cast(16739493649928310215, json BINARY), test.tceb7972c.col_17), not(istrue_with_null(json_contains(test.tceb7972c.col_17, cast("6019730272580550835", json BINARY)))))
  │   └─TableFullScan_14  6.00  cop[tikv] table:tceb7972c keep order:false, stats:partial[col_17:missing]
  └─Projection_17(Probe)  10000.00  root    cast(test.t61a85298.col_71, double BINARY)->Column#19
    └─TableReader_19  10000.00  root    data:TableFullScan_18
      └─TableFullScan_18  10000.00  cop[tikv] table:t61a85298 keep order:false, stats:pseudo

• No need to test

  • I checked and no code files have been changed.

Side effects

• Performance regression: Consumes more CPU
• Performance regression: Consumes more Memory
• Breaking backward compatibility

Documentation

• Affects user behaviors
• Contains syntax changes
• Contains variable changes
• Contains experimental features
• Changes MySQL compatibility

Release note

Please refer to Release Notes Language Style Guide to write a quality release note.

None

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 56.7427%. Comparing base (2b54db5) to head (4382573).
Report is 38 commits behind head on master.

Additional details and impacted files

@@                Coverage Diff                @@
##             master     #55842         +/-   ##
=================================================
- Coverage   72.8695%   56.7427%   -16.1269%     
=================================================
  Files          1588       1759        +171     
  Lines        443554     647686     +204132     
=================================================
+ Hits         323216     367515      +44299     
- Misses       100456     255197     +154741     
- Partials      19882      24974       +5092     

Flag	Coverage Δ
integration	38.6513% <100.0000%> (?)
unit	72.0245% <100.0000%> (+0.0652%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Components	Coverage Δ
dumpling	52.9567% <ø> (ø)
parser	∅ <ø> (∅)
br	51.8079% <ø> (+6.3936%)	⬆️

[hawkingrei]

/retest

[time-and-fate]

I think the root cause here is that the global index (idx_10 and idx_9) here is not analyzed actually, but they are marked as analyzed (according to ColAndIdxExistenceMap) and fully loaded (according to StatsLoadedStatus).
So we will try to use this here, then the idxStats.TotalRowCount() is 0 in GetScaledRealtimeAndModifyCnt(), causing the problem.

[hawkingrei]

And I think we can't close the issue in this PR, right? Since the wrong result is not fixed.

The wrong result has been fixed.

BTW, col_17 is the JSON column. We cannot analyze it by default. so the index is related with col_17 will be without stats.

[time-and-fate]

MV index has stats, but idx_10 and idx_9 don't have stats here. It's likely because they are global index. I think the root cause is still that the not analyzed stats are marked analyzed and fully loaded.
For a temporary fix, I think we can add a check for idxStats.TotalRowCount() == 0 in GetScaledRealtimeAndModifyCnt().

[hawkingrei]

MV index has stats, but idx_10 and idx_9 don't have stats here. It's likely because they are global index. I think the root cause is still that the not analyzed stats are marked analyzed and fully loaded. For a temporary fix, I think we can add a check for idxStats.TotalRowCount() == 0 in GetScaledRealtimeAndModifyCnt().

I have updated it again.

[hawkingrei]

/retest

[time-and-fate]

Please restore tidb_enable_global_index after the tests.

[ti-chi-bot]

[APPROVALNOTIFIER] This PR is APPROVED

This pull-request has been approved by: qw4990, time-and-fate

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

The pull request process is described here

Needs approval from an approver in each of these files:

• OWNERS [qw4990,time-and-fate]
• pkg/statistics/OWNERS [qw4990,time-and-fate]

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment

[ti-chi-bot]

[LGTM Timeline notifier]

Timeline:

• 2024-09-05 06:18:51.084216555 +0000 UTC m=+511655.602269493: ☑️ agreed by qw4990.
• 2024-09-06 10:45:07.042135536 +0000 UTC m=+7576.782559475: ☑️ agreed by time-and-fate.

[hawkingrei]

/hold

[hawkingrei]

/unhold

[hawkingrei]

/retest

[hawkingrei]

/cherrypick release-8.1
/cherrypick release-7.5

[ti-chi-bot]

@hawkingrei: new pull request created to branch release-7.5: #55923.

In response to this:

/cherrypick release-8.1
/cherrypick release-7.5

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 ti-community-infra/tichi repository.

[ti-chi-bot]

@hawkingrei: new pull request created to branch release-8.1: #55924.

In response to this:

/cherrypick release-8.1
/cherrypick release-7.5

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 ti-community-infra/tichi repository.