Data lost when pause and resume changefeed while executing drop table DDL.

[asddongmen]

What did you do?

All steps are in this integration case: https://github.com/pingcap/tiflow/blob/master/tests/integration_tests/batch_update_to_no_batch/run.sh

1. create changefeed c1.
2. drop table t1 in upstream
3. recover table t1 in upstream
4. insert a row r1 into t1
5. pause c1
6. resume c1

What did you expect to see?

row r1 can be replicated to downstream

What did you see instead?

there is no r1 in downstream

Versions of the cluster

Upstream TiDB cluster version (execute SELECT tidb_version(); in a MySQL client):

(paste TiDB cluster version here)

Upstream TiKV version (execute tikv-server --version):

(paste TiKV version here)

TiCDC version (execute cdc version):

master

[asddongmen]

When a changefeed is paused while there is a DDL executing, DDL's executing would be canceld, and ddlEventCache would not be clean. While changefeed resume in auto retring, the DDL in ddlEventCache will be sent to downstream directly but not be applied to the schema of the changefeed.
So, changefeed will loss the tables, which lead to data lost.