DDL maybe block forever if owner is restart or progress the schema change unexpectedly

[wjhuang2016]

Bug Report

Please answer these questions before submitting your issue. Thanks!

1. Minimal reproduce step (Required)

Given two tables, namely t1, t2.
And two TiDB servers, tidb1(ddl owner), tidb2.
Note: It may not reproduce for 2 tables/DDLs, you can extend to 3 tables/DDLs.

1. connect to tidb 2. Do:

begin;
select * from t1, t2;

2. connect to tidb 2. Do:

alter table t1 add column b int;

3. connect to tidb 2. Do:

alter table t2 add column b int;

4. restart/close/kill TiDB1(owner)
5. commit the transaction in step1 session.

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

Two DDL success finally.

3. What did you see instead (Required)

The first DDL success finally, but the second DDL block forever(even with restarting the whole cluster),

4. What is your TiDB version? (Required)

6.5, (7.1, master) No blocking, but progress the schema change unexpectedly .