TestMultiSchemaModifyColumn shows index inconsistency

[mjonss]

Bug Report

Please answer these questions before submitting your issue. Thanks!

1. Minimal reproduce step (Required)

Check out c3ea51e , make failpoint-enable and run TestMultiSchemaModifyColumn.

2. What did you expect to see? (Required)

Test passing

3. What did you see instead (Required)

If adding admin check table t it fails with:
[admin:8223]data inconsistency in table: t, index: uk_b, handle: 10, index-values:"" != record-values:"handle: 10, values: [KindNull ]"
otherwise it fails with:

[2025/03/25 23:10:52.587 +01:00] [INFO] [multi_domain_test.go:1209] [HaveEntriesForTableIndex] [key=7480000000000000705f69800000000000000101203031312e353000fe] [value=08800001000000020800203031312e353020000000000000000b]
[2025/03/25 23:10:52.587 +01:00] [INFO] [multi_domain_test.go:1214] [HaveEntriesForTableIndex] [tableID=112] [indexID=1] [count=1]
    multi_domain_test.go:902: 
        	Error Trace:	/repos/tidb/pkg/ddl/tests/partition/multi_domain_test.go:902
        	            				/repos/tidb/pkg/ddl/tests/partition/multi_domain_test.go:1140
        	            				/repos/tidb/pkg/ddl/tests/partition/multi_domain_test.go:935
        	            				/repos/tidb/pkg/ddl/tests/partition/multi_domain_test.go:643
        	Error:      	Should be false
        	Test:       	TestMultiSchemaModifyColumn
        	Messages:   	Index id range [1,2) for table id 112 has still entries!

4. What is your TiDB version? (Required)

Similar to current master.

tidb_version(): Release Version: v9.0.0-alpha-506-gc3ea51e172
Edition: Community
Git Commit Hash: c3ea51e172a0f2688bf7a3b16b78fff38971efba
Git Branch: non-clustered-reorg-duplicate-_tidb_rowid-fix
UTC Build Time: 2025-03-25 21:57:39
GoVersion: go1.24.0
Race Enabled: false
Check Table Before Drop: false
Store: unistore

[lance6716]

also cc @wjhuang2016

[wjhuang2016]

master works fine.

[mjonss]

Another way to reproduce on current master, apply this diff, make failpoint-enable and run the test:

diff --git a/pkg/ddl/tests/partition/multi_domain_test.go b/pkg/ddl/tests/partition/multi_domain_test.go
index d743e2cc24..af3eced0e7 100644
--- a/pkg/ddl/tests/partition/multi_domain_test.go
+++ b/pkg/ddl/tests/partition/multi_domain_test.go
@@ -634,6 +634,7 @@ func TestMultiSchemaModifyColumn(t *testing.T) {
                default:
                        require.Failf(t, "unhandled schema state '%s'", schemaState)
                }
+               tkO.MustExec(`admin check table t`)
        }
        runMultiSchemaTest(t, createSQL, alterSQL, initFn, nil, loopFn)
 }

It then fails with:

[admin:8223]data inconsistency in table: t, index: uk_b, handle: 10, index-values:"" != record-values:"handle: 10, values: [KindNull <nil>]"

[CbcWestwolf]

The root cause is that when the ddl owner updates the state to public and the non-owner remains in write reorg, the non-owner uses the old index in the old schema, whose data is deleted by delete range. This code branch prints some debug information https://github.com/CbcWestwolf/tidb/tree/fix_issue60264