*: activate txn for query on infoschema tables

[tiancaiamao]

What problem does this PR solve?

Issue Number: close #57874

Problem Summary:

The GC should be blocked by query.

What changed and how does it work?

There are two places that the transaction should blcok GC:

• The query "SELECT COUNT(1) as c FROM INFORMATION_SCHEMA.TABLES WHERE TABLE_TYPE='SEQUENCE'" itself
• The internal transaction meta.ListTables invoked by SchemaTableInfos() API

This commit just fix the first one. It should be enough to fix issue 57874.
The root cause is that whe we query on infoschema tables, the transaction is not activated.
So the session process info would not be collected to block GC.

The second issue is caused by the fact that the meta API call does not block GC.
I don't take investigation on it further more, meta API is limited used.

Check List

Tests

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

Modify the code and build tidb, check no more "GC life time is shorter than transaction duration" error.

diff --git a/pkg/planner/core/memtable_infoschema_extractor.go b/pkg/planner/core/memtable_infoschema_extractor.go
index c527dde968..34cabb3677 100644
--- a/pkg/planner/core/memtable_infoschema_extractor.go
+++ b/pkg/planner/core/memtable_infoschema_extractor.go
@@ -15,6 +15,7 @@
 package core
 
 import (
+       "time"
        "bytes"
        "context"
        "fmt"
@@ -184,6 +185,7 @@ func (e *InfoSchemaBaseExtractor) ListSchemasAndTables(
                return findTableAndSchemaByName(ctx, is, schemas, tableNames)
        }
        schemas := e.ListSchemas(is)
+       time.Sleep(12 * time.Minute)
        return listTablesForEachSchema(ctx, e, is, schemas)
 }

mysql> set @@tidb_mem_quota_query = 12*1024*1024*1024;
Query OK, 0 rows affected (0.00 sec)

mysql> select count(tidb_table_id) from information_schema.tables where table_name != 'test';
+----------------------+
| count(tidb_table_id) |
+----------------------+
|              1000818 |
+----------------------+
1 row in set (12 min 28.84 sec)

• 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

[tiprow]

Hi @tiancaiamao. Thanks for your PR.

PRs from untrusted users cannot be marked as trusted with /ok-to-test in this repo meaning untrusted PR authors can never trigger tests themselves. Collaborators can still trigger tests on the PR using /test all.

I understand the commands that are listed here.

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-sigs/prow repository.

[tiancaiamao]

/hold

[tiancaiamao]

The second case is still risky after this fix.

If we take a latest infoschema, and use it, continue for 10min (or maybe less)
The GC error may still happen because the internal txn does not block GC.

[tiancaiamao]

/unhold

[codecov]

Codecov Report

Attention: Patch coverage is 75.00000% with 8 lines in your changes missing coverage. Please review.

Project coverage is 73.6233%. Comparing base (5598c11) to head (6e4ab15).
Report is 6 commits behind head on master.

Additional details and impacted files

@@               Coverage Diff                @@
##             master     #57937        +/-   ##
================================================
+ Coverage   73.1836%   73.6233%   +0.4396%     
================================================
  Files          1671       1671                
  Lines        460712     461624       +912     
================================================
+ Hits         337166     339863      +2697     
+ Misses       102825     101035      -1790     
- Partials      20721      20726         +5     

Flag	Coverage Δ
integration	43.2697% <75.0000%> (?)
unit	72.3051% <75.0000%> (-0.0172%)	⬇️

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

Components	Coverage Δ
dumpling	52.6910% <ø> (-0.0764%)	⬇️
parser	∅ <ø> (∅)
br	46.0134% <ø> (-0.0121%)	⬇️

[lance6716]

Can you find the place that min TS of active txn ? We can write a testkit test to check the minTS really takes effect with this PR

[ti-chi-bot]

[APPROVALNOTIFIER] This PR is APPROVED

This pull-request has been approved by: lance6716, wjhuang2016

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 [lance6716,wjhuang2016]

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:

• 2024-12-03 08:53:28.343175453 +0000 UTC m=+1144995.962829968: ☑️ agreed by lance6716.
• 2024-12-03 14:21:45.598312086 +0000 UTC m=+1164693.217966596: ☑️ agreed by wjhuang2016.

[wjhuang2016]

/hold wait for adding comment.

[tiancaiamao]

Can you find the place that min TS of active txn ? We can write a testkit test to check the minTS really takes effect with this PR

It's here @lance6716

tidb/pkg/domain/infosync/info.go

Lines 760 to 765 in d021648

	func (is *InfoSyncer) ReportMinStartTS(store kv.Storage) {
	sm := is.GetSessionManager()
	if sm == nil {
	return
	}
	pl := sm.ShowProcessList()

The start ts is collected by geting ProcessInfo.
Actually, I tired to add some tesr to cover that executing "select * from information_schema.xxx", the start ts of session process info is non-zero.
But I find the failpoint is ugly and the cover check looks like "prove my mother is my mother", so I give up.

func TestInfoschemaTableActivateTxn(t *testing.T) {
	store := testkit.CreateMockStore(t)
	tk := testkit.NewTestKit(t, store)

	// This test checks that query on infoschema tables should activate txn.
	testfailpoint.EnableCall(t, "github.com/pingcap/tidb/pkg/executor/inspectSession", func(sctx sessionctx.Context) {
		pi := sctx.ShowProcess()
		require.True(t, pi.CurTxnStartTS > 0)
		fmt.Println("============================================================")
	})


	for _, sql := range []string{
		"select count(tidb_table_id) > 0 from information_schema.tables where table_schema != 'test'",
		"select count(*) from information_schema.sequences",
	}{
		tk.MustQuery(sql)
	}

	require.True(t, called)
}

(If e.Ctx().Txn(true) is called, start ts is non-zero, otherwise it' zero)

[tiancaiamao]

/unlock

[tiancaiamao]

/unhold

[ti-chi-bot]

In response to a cherrypick label: new pull request created to branch release-8.5: #57951.

[tiancaiamao]

Create an issue #57952 to trace the second problem.
That one is not blocking, I list as en enhancement, as most of the cases should have been avoided by this commit.