ddl: add restore deleted table

[crazycs520]

What problem does this PR solve?

Add admin restore table by job job_id to restore the deleted table.
related proposal: #7383
related parser PR: pingcap/parser#85, pingcap/parser#172

What is changed ?

Add a new ddl type job: restore table to do restore.
the function of restore table almost as same as create table ddl.
But still a little bit different:

• restore table don't do split table region.
• restore table will rebase the table auto id with the original table autoID.

restore table usage

Scenario 1

If you drop table by mistake, you can use admin restore table [table_name] to restore the table.

Eg:

mysql> use test
You are now connected to database "test" as user "root"
mysql> create table t (a int)
mysql> insert into t values (1);
mysql> select a from t;
+---+
| a |
+---+
| 1 |
+---+
mysql> drop table t;
mysql> select a from t;
(1146, u"Table 'test.t' doesn't exist")
mysql> admin restore table test.t;
mysql> select a from t;
+---+
| a |
+---+
| 1 |
+---+
mysql> insert into t values (2);
mysql> select a from t;
+---+
| a |
+---+
| 1 |
| 2 |
+---+

Scenario 2

If you drop table t by mistable, then create a new table t, then drop the new table t, now, if you use admin restore table t, It will restore the new table t, not the ole table t . But how to restore the old table t ? You can use admin restore table by job [drop_table_job_ID] .

Eg:

mysql> use test
You are now connected to database "test" as user "root"
mysql> select * from t;
+---+
| a |
+---+
| 1 |
+---+
mysql> drop table t;
mysql> create table t (a int,b varchar(10));
mysql> insert into t values (1,"new t");
mysql> select * from t;
+---+-------+
| a | b     |
+---+-------+
| 1 | new t |
+---+-------+
mysql> drop table t;
mysql> admin restore table test.t;
mysql> select * from t;
+---+-------+
| a | b     |
+---+-------+
| 1 | new t |
+---+-------+
mysql> admin show ddl jobs;
+--------+---------+------------+---------------+--------------+-----------+----------+-----------+-----------------------------------+--------+
| JOB_ID | DB_NAME | TABLE_NAME | JOB_TYPE      | SCHEMA_STATE | SCHEMA_ID | TABLE_ID | ROW_COUNT | START_TIME                        | STATE  |
+--------+---------+------------+---------------+--------------+-----------+----------+-----------+-----------------------------------+--------+
| 73     | test    | t          | restore table | public       | 1         | 70       | 0         | 2019-01-13 14:55:59.016 +0800 CST | synced |
| 72     | test    | t          | drop table    | none         | 1         | 70       | 0         | 2019-01-13 14:55:54.116 +0800 CST | synced |
| 71     | test    | t          | create table  | public       | 1         | 70       | 0         | 2019-01-13 14:55:17.414 +0800 CST | synced |
| 69     | test    |            | drop table    | none         | 1         | 60       | 0         | 2019-01-13 14:55:03.262 +0800 CST | synced |
| 68     | test    |            | add index     | public       | 1         | 60       | 0         | 2019-01-13 14:54:28.415 +0800 CST | synced |
| 67     | test    |            | add index     | public       | 1         | 60       | 0         | 2019-01-13 14:54:24.563 +0800 CST | synced |
| 66     | test    |            | add index     | public       | 1         | 60       | 0         | 2019-01-13 14:54:20.611 +0800 CST | synced |
| 65     | test    |            | add index     | public       | 1         | 60       | 0         | 2019-01-13 14:54:10.915 +0800 CST | synced |
| 64     | test    |            | add index     | public       | 1         | 60       | 0         | 2019-01-13 14:54:05.511 +0800 CST | synced |
| 63     | test    |            | add index     | public       | 1         | 60       | 0         | 2019-01-13 14:54:01.411 +0800 CST | synced |
+--------+---------+------------+---------------+--------------+-----------+----------+-----------+-----------------------------------+--------+
10 rows in set
mysql> admin restore table by job 69; /* if you not sure which job drop the old table, you can also find this in tidb log file. */
(1050, u"Table 't' already exists")  /* because there is already have a table name `t`, so have to rename the table first. */
mysql> rename table t to t_new;
mysql> admin restore table by job 69;
mysql> show tables;
+----------------+
| Tables_in_test |
+----------------+
| t              |
| t_new          |
+----------------+
mysql> select * from t;
+---+
| a |
+---+
| 1 |
+---+
mysql> select * from t_new;
+---+-------+
| a | b     |
+---+-------+
| 1 | new t |
+---+-------+

Watch out

If you drop table t multiple time, then use admin recover table by job [ID] with the old drop job ID, there will have some problem when you write.

Eg:

mysql> use test
You are now connected to database "test" as user "root"
mysql> insert into t set b=1;
mysql> select * from t;
+---+---+
| a | b |
+---+---+
| 1 | 1 |
+---+---+
mysql> drop table t;
mysql> admin restore table t;
mysql> insert into t set b=2;
mysql> select * from t;
+-------+---+
| a     | b |
+-------+---+
| 1     | 1 |
| 30001 | 2 |
+-------+---+
mysql> drop table t;
mysql> admin show ddl jobs;
+--------+---------+------------+---------------+--------------+-----------+----------+-----------+-----------------------------------+--------+
| JOB_ID | DB_NAME | TABLE_NAME | JOB_TYPE      | SCHEMA_STATE | SCHEMA_ID | TABLE_ID | ROW_COUNT | START_TIME                        | STATE  |
+--------+---------+------------+---------------+--------------+-----------+----------+-----------+-----------------------------------+--------+
| 106    | test    |            | drop table    | none         | 1         | 94       | 0         | 2019-01-13 15:21:27.016 +0800 CST | synced |
| 105    | test    |            | restore table | public       | 1         | 94       | 0         | 2019-01-13 15:19:37.564 +0800 CST | synced |
| 104    | test    |            | drop table    | none         | 1         | 94       | 0         | 2019-01-13 15:19:29.318 +0800 CST | synced |
| 103    | test    |            | drop index    | none         | 1         | 94       | 0         | 2019-01-13 15:18:56.266 +0800 CST | synced |
| 102    | test    |            | drop index    | none         | 1         | 94       | 0         | 2019-01-13 15:18:50.815 +0800 CST | synced |
| 101    | test    |            | drop index    | none         | 1         | 94       | 0         | 2019-01-13 15:18:47.715 +0800 CST | synced |
| 100    | test    |            | drop index    | none         | 1         | 94       | 0         | 2019-01-13 15:18:45.364 +0800 CST | synced |
| 99     | test    |            | add index     | public       | 1         | 94       | 0         | 2019-01-13 15:18:25.263 +0800 CST | synced |
| 98     | test    |            | add index     | public       | 1         | 94       | 0         | 2019-01-13 15:18:22.063 +0800 CST | synced |
| 97     | test    |            | add index     | public       | 1         | 94       | 0         | 2019-01-13 15:18:10.763 +0800 CST | synced |
+--------+---------+------------+---------------+--------------+-----------+----------+-----------+-----------------------------------+--------+
mysql> admin restore table by job 104;
mysql> select * from t;
+-------+---+
| a     | b |
+-------+---+
| 1     | 1 |
| 30001 | 2 |
+-------+---+
mysql> insert into t set b=3;
mysql> select * from t;
+-------+---+
| a     | b |
+-------+---+
| 1     | 1 |
| 30001 | 3 |  /* Watch out: miss record b=2, This because record b=2 is covered by record b=3.*/
+-------+---+

summary

Normally recommend use admin restore table [table_name] to restore table. admin restore table by job [ID] is only for the Scenario 2.

How it works ?

This is done through the following steps:

1. Get the history ddl job by job_id.
2. Use the drop table job's StartTS to get a snapshot infoSchema before drop table.
3. Find the table meta by table_id in snapshot infoSchema, if not found, return ErrTableIDNotExists.
4. Using the Found table meta to create table( recover table ). the table_id is still the original table id. Return if encounter any error.
5. Remove dropped table DDL job from gc_delete_range table by drop table job's id and table_id.

But Change Settings and undo Settings will make the problem more complicated.

Check List

Tests

• Unit test
• Manual test

Code changes

• Has exported function/method change
• Has exported variable/fields change
• Has interface methods change

Side effects

• Increased code complexity

Related changes

---

This change is 

[crazycs520]

@winkyao @zimulala @ciscoxll PTAL.

[crazycs520]

/run-all-tests

[crazycs520]

/run-unit-test

[crazycs520]

@zimulala @winkyao @ciscoxll PTAL

[zyguan]

Hi, @crazycs520 , ActionRestoreTable is undefined. you may need to bump parser first.

[zhouqiang-cl]

/rebuild

[crazycs520]

db_partition_test

[crazycs520]

/run-all-tests

[crazycs520]

/run-all-tests

[ciscoxll]

LGTM.

[crazycs520]

/run-all-tests

[crazycs520]

/run-all-tests

[crazycs520]

@zimulala PTAL again.

[crazycs520]

/run-all-tests

[crazycs520]

/run-unit-test

[zimulala]

Do we need to consider ServerStatusInTrans like other DDLs?

[zimulala]

LGTM

[crazycs520]

/run-all-tests

[crazycs520]

/run-unit-test