fast DDL failed with [Lightning:Restore:ErrOpenDuplicateDB]open duplicate db error: open /tmp/tidb/tmp_ddl-4000/595/duplicates/MANIFEST-000001: no such file or directory

[lance6716]

Bug Report

Please answer these questions before submitting your issue. Thanks!

1. Minimal reproduce step (Required)

concurrently run multiple fast ADD INDEX

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

no error

3. What did you see instead (Required)

[Lightning:Restore:ErrOpenDuplicateDB]open duplicate db error: open /tmp/tidb/tmp_ddl-4000/595/duplicates/MANIFEST-000001: no such file or directory

4. What is your TiDB version? (Required)

[lance6716]

I think the problem is when the first DDL job finished one schema state transition, it will remove itself from runningJobs

tidb/pkg/ddl/job_table.go

Lines 416 to 423 in a9e3398

	// delivery2Worker owns the worker, need to put it back to the pool in this function.
	func (d *ddl) delivery2Worker(wk *worker, pool *workerPool, job *model.Job) {
	injectFailPointForGetJob(job)
	d.runningJobs.add(job)
	d.wg.Run(func() {
	metrics.DDLRunningJobCount.WithLabelValues(pool.tp().String()).Inc()
	defer func() {
	d.runningJobs.remove(job)

Then the second DDL job have chance to be run. Meanwhile, the first job still can continue the next state transition because it has processing set to 1

tidb/pkg/ddl/job_table.go

Lines 121 to 126 in a9e3398

	// The job has already been picked up, just return to continue it.
	if isJobProcessing {
	return &job, nil
	}
	b, err := filter(&job)

So the second job may remove the first job's sort directory if the first job does not add itself to runningJobs timely in the next state transition.

[lance6716]

closed by #52503