DDL handling fails when changefeed starts at FinishTs of a schema change

[liuzix]

Bug Report

Please answer these questions before submitting your issue. Thanks!

1. What did you do? If possible, provide a recipe for reproducing the error.

• Truncate some tables, and while the DDLs are still being executed, kill the owner.

2. What did you expect to see?

• The cluster can recover and replication continues

3. What did you see instead?

• CDC:ErrSchemaStorageTableMiss.

4. Versions of the cluster

   • TiCDC version (execute cdc version):

     v5.2.0
     

5. Cause

The TS used for snapshot read from TiKV is not correctly set. https://github.com/pingcap/ticdc/blob/04fa75b2ae194e481e8a82c0cd12bf7d8ca2a249/cdc/owner/schema.go#L44

        var meta *timeta.Meta
	if kvStorage != nil {
		var err error
		meta, err = kv.GetSnapshotMeta(kvStorage, startTs)  // wrong! should be startTs-1
		if err != nil {
			return nil, errors.Trace(err)
		}
	}
	// We do a snapshot read of the metadata from TiKV at (startTs-1) instead of startTs,
	// because the DDL puller might send a DDL at startTs, which would cause schema conflicts if
	// the DDL's result is already contained in the snapshot.
	schemaSnap, err := entry.NewSingleSchemaSnapshotFromMeta(meta, startTs-1, config.ForceReplicate)
	if err != nil {
		return nil, errors.Trace(err)
	}