subtask can't be re-assigned back when tidb node restart

[seiya-annie]

Bug Report

Please answer these questions before submitting your issue. Thanks!

1. Minimal reproduce step (Required)

1. run import into command, data in s3, and global-sort mode
2. check subtask assigned to different tidb node
3. rolling restart tidb
4. after tidb recoveried, check subtask

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

runnable subtasks should be assigned to different tidb node

3. What did you see instead (Required)

all subtask are in one tidb node

4. What is your TiDB version? (Required)

master

[seiya-annie]

before restart

mysql> select id, state, step, exec_id, cast(error as char) from mysql.tidb_background_subtask where task_key=(select id from mysql.tidb_global_task where task_key='ImportInto/30005' );
+------+---------+------+---------------------------------------------------------------------------------------+---------------------+
| id   | state   | step | exec_id                                                                               | cast(error as char) |
+------+---------+------+---------------------------------------------------------------------------------------+---------------------+
| 1733 | running |    3 | [tc-tidb-4.tc](http://tc-tidb-4.tc/)-tidb-peer.sql-global-sort-and-mvindex-testing-tps-2760376-1-696.svc:4000 | NULL                |
| 1734 | running |    3 | [tc-tidb-1.tc](http://tc-tidb-1.tc/)-tidb-peer.sql-global-sort-and-mvindex-testing-tps-2760376-1-696.svc:4000 | NULL                |
+------+---------+------+---------------------------------------------------------------------------------------+---------------------+
2 rows in set (0.01 sec)

after restart

MySQL [(none)]> select id, state, step, exec_id, cast(error as char) from mysql.tidb_background_subtask where task_key=(select id from mysql.tidb_global_task where task_key='ImportInto/30005');
+------+---------+------+---------------------------------------------------------------------------------------+---------------------+
| id   | state   | step | exec_id                                                                               | cast(error as char) |
+------+---------+------+---------------------------------------------------------------------------------------+---------------------+
| 1733 | running |    3 | [tc-tidb-2.tc](http://tc-tidb-2.tc/)-tidb-peer.sql-global-sort-and-mvindex-testing-tps-2760376-1-696.svc:4000 | NULL                |
| 1734 | pending |    3 | [tc-tidb-2.tc](http://tc-tidb-2.tc/)-tidb-peer.sql-global-sort-and-mvindex-testing-tps-2760376-1-696.svc:4000 | NULL                |
+------+---------+------+---------------------------------------------------------------------------------------+---------------------+
2 rows in set (0.01 sec)

[ywqzzy]

Not a bug.
We can mark it as an enhancement issue since the task behavior will not changed.
We can do some optimization to make the subtasks executor load balanced.