lightning parallel import single table with duplication data fail when duplicate-resolution = record

[seiya-annie]

Bug Report

Please answer these questions before submitting your issue. Thanks!

1. Minimal reproduce step (Required)

1. Launch 2 lightning instance to do parallel local backend import data for single data, with duplicate data, specify duplicate-resolution = record
2. Lightning should succeed with notification to users to let them know there is duplicate data

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

1. Lighting import should succeed (crc32 check skipped as data might be inconsistent)
2. 3. Duplicate rows numbers check should succeed

3. What did you see instead (Required)

[2022/11/29 18:43:32.126 +00:00] [ERROR] [main.go:103] ["tidb lightning encountered error stack info"] [error="failed to record conflict errors: [xeval:8221]invalid key - "7480000000000002b55f6980000000000000010419a6340000000000013138343939323234ff0000000000000000f7016c69505a00000000fb""] [errorVerbose="[xeval:8221]invalid key - "7480000000000002b55f6980000000000000010419a6340000000000013138343939323234ff0000000000000000f7016c69505a00000000fb"\ngithub.com/pingcap/errors.AddStack\n\t/go/pkg/mod/github.com/pingcap/errors@v0.11.5-0.20220729040631-518f63d66278/errors.go:174\ngithub.com/pingcap/errors.(*Error).GenWithStack\n\t/go/pkg/mod/github.com/pingcap/errors@v0.11.5-0.20220729040631-518f63d66278/normalize.go:155\ngithub.com/pingcap/tidb/tablecodec.DecodeRowKey\n\t/home/jenkins/agent/workspace/build-common/go/src/github.com/pingcap/br/tablecodec/tablecodec.go:283\ngithub.com/pingcap/tidb/br/pkg/lightning/backend/kv.(*TableKVDecoder).DecodeHandleFromRowKey\n\t/home/jenkins/agent/workspace/build-common/go/src/github.com/pingcap/br/br/pkg/lightning/backend/kv/kv2sql.go:42\ngithub.com/pingcap/tidb/br/pkg/lightning/backend/local.(*DuplicateManager).RecordDataConflictError\n\t/home/jenkins/agent/workspace/build-common/go/src/github.com/pingcap/br/br/pkg/lightning/backend/local/duplicate.go:444\ngithub.com/pingcap/tidb/br/pkg/lightning/backend/local.(*DuplicateManager).processRemoteDupTaskOnce.func1.1\n\t/home/jenkins/agent/workspace/build-common/go/src/github.com/pingcap/br/br/pkg/lightning/backend/local/duplicate.go:780\ngithub.com/pingcap/tidb/br/pkg/lightning/backend/local.(*DuplicateManager).processRemoteDupTaskOnce.func1\n\t/home/jenkins/agent/workspace/build-common/go/src/github.com/pingcap/br/br/pkg/lightning/backend/local/duplicate.go:788\ngithub.com/pingcap/tidb/br/pkg/utils.(*WorkerPool).Apply.func1\n\t/home/jenkins/agent/workspace/build-common/go/src/github.com/pingcap/br/br/pkg/utils/worker.go:58\nruntime.goexit\n\t/usr/local/go/src/runtime/asm_amd64.s:1594\nfailed to record conflict errors"]

4. What is your TiDB version? (Required)

[2022/11/29 18:41:59.964 +00:00] [INFO] [info.go:49] ["Welcome to TiDB-Lightning"] [release-version=v6.5.0-alpha] [git-hash=9689b4763d2705f1dc1308b4e4bec257e71d391a] [git-branch=heads/refs/tags/v6.5.0-alpha] [go-version=go1.19.3] [utc-build-time="2022-11-29 11:09:16"] [race-enabled=false]

[seiya-annie]

lightning.log

[dsdashun]

Here's the root cause: The key is an index key, but the dup-detector logic treated it as a row record key. The logic is here:

tidb/br/pkg/lightning/backend/local/duplicate.go

Lines 779 to 783 in 9689b47

	if task.indexInfo == nil {
	err = m.RecordDataConflictError(ctx, stream)
	} else {
	err = m.RecordIndexConflictError(ctx, stream, task.tableID, task.indexInfo)
	}

It should have an indexInfo and jump into the else clause, but it actually jumped into the RecordDataConflictError branch.
When constructing the task struct using (*DuplicateManager) buildDupTasks(), now it abstracted the appending task operation as a single function

tidb/br/pkg/lightning/backend/local/duplicate.go

Line 556 in 9689b47

putToTaskFunc := func(ranges []tidbkv.KeyRange) {

However, when appending the index dup-detect task, it should add the indexInfo, but unfortunately, the abstracted function didn't do that:

tidb/br/pkg/lightning/backend/local/duplicate.go

Lines 562 to 565 in 9689b47

	tasks = append(tasks, dupTask{
	KeyRange: r,
	tableID: tid,
	})