The lock resolve is slow for p-dml & the progress log of range task may be omitted

[you06]

Phenomenon

From the metrics, the resolve-lock stage of p-dml is very slow.

And the logs does not log the progress every 30 seconds as expected, the range task runner only print the following log.

[2025/02/18 02:25:07.305 +00:00] [INFO] [range_task.go:167] ["range task started"] [name=pipelined-dml-commit-456089876295581703] [startKey=74800000000000007e5f69800000000000000103800000000000000103800000000185fa98] [endKey=74800000000000007e5f728000000005f5e100] [concurrency=8]
[2025/02/18 03:09:28.607 +00:00] [INFO] [range_task.go:276] ["range task finished"] [name=pipelined-dml-commit-456089876295581703] [startKey=74800000000000007e5f69800000000000000103800000000000000103800000000185fa98] [endKey=74800000000000007e5f728000000005f5e100] ["cost time"=44m21.302468224s] ["completed regions"=201]

Analysis

The progress log is print here:

client-go/txnkv/rangetask/range_task.go

Lines 207 to 217 in 456c3af

	select {
	case <-statLogTicker.C:
	logutil.Logger(ctx).Info("range task in progress",
	zap.String("name", s.identifier),
	zap.String("startKey", kv.StrKey(startKey)),
	zap.String("endKey", kv.StrKey(endKey)),
	zap.Int("concurrency", s.concurrency),
	zap.Duration("cost time", time.Since(startTime)),
	zap.Int("completed regions", s.CompletedRegions()))
	default:
	}

However, the progress log may not be printed if there are not such a lot of tasks, and it's waiting here.

client-go/txnkv/rangetask/range_task.go

Line 261 in 456c3af

wg.Wait()