tidb_decode_key doesn't work for clustered index

[kolbe]

Bug Report

Please answer these questions before submitting your issue. Thanks!

1. Minimal reproduce step (Required)

set tidb_enable_clustered_index=ON;
create table t1 ( `a` int(10) unsigned NOT NULL, b int unsigned not null, PRIMARY KEY (a,b) /*T![clustered_index] CLUSTERED */);
split table t1 between (0,0) and (3000,0) regions 10;
select start_key, tidb_decode_key(start_key) from information_schema.tikv_region_status where table_name='t1';

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

3. What did you see instead (Required)

5.7.25-TiDB-v4.0.0-beta.2-2292-g4218f2836-dirty (root) [test]> set tidb_enable_clustered_index=ON;
Query OK, 0 rows affected (0.000 sec)

create table t1 ( `a` int(10) unsigned NOT NULL, b int unsigned not null, PRIMARY KEY (a,b) /*T![clustered_index] CLUSTERED */);
split table t1 between (0,0) and (3000,0) regions 10;
select start_key, tidb_decode_key(start_key) from information_schema.tikv_region_status where table_name='t1';
5.7.25-TiDB-v4.0.0-beta.2-2292-g4218f2836-dirty (root) [test]> create table t1 ( `a` int(10) unsigned NOT NULL, b int unsigned not null, PRIMARY KEY (a,b) /*T![clustered_index] CLUSTERED */);
Query OK, 0 rows affected (0.078 sec)

5.7.25-TiDB-v4.0.0-beta.2-2292-g4218f2836-dirty (root) [test]> split table t1 between (0,0) and (3000,0) regions 10;
+--------------------+----------------------+
| TOTAL_SPLIT_REGION | SCATTER_FINISH_RATIO |
+--------------------+----------------------+
|                  9 |                    1 |
+--------------------+----------------------+
1 row in set (1.367 sec)

5.7.25-TiDB-v4.0.0-beta.2-2292-g4218f2836-dirty (root) [test]> select start_key, tidb_decode_key(start_key) from information_schema.tikv_region_status where table_name='t1';
+--------------------------------------------------------------------------+--------------------------------------------------------------------------+
| start_key                                                                | tidb_decode_key(start_key)                                               |
+--------------------------------------------------------------------------+--------------------------------------------------------------------------+
| 7480000000000000FF665F720400000000FF000004B004000000FF0000000000000000F9 | 7480000000000000FF665F720400000000FF000004B004000000FF0000000000000000F9 |
| 7480000000000000FF665F720400000000FF0000012C04000000FF0000000000000000F9 | 7480000000000000FF665F720400000000FF0000012C04000000FF0000000000000000F9 |
| 7480000000000000FF665F720400000000FF0000025804000000FF0000000000000000F9 | 7480000000000000FF665F720400000000FF0000025804000000FF0000000000000000F9 |
| 7480000000000000FF665F720400000000FF000005DC04000000FF0000000000000000F9 | 7480000000000000FF665F720400000000FF000005DC04000000FF0000000000000000F9 |
| 7480000000000000FF665F720400000000FF0000083404000000FF0000000000000000F9 | 7480000000000000FF665F720400000000FF0000083404000000FF0000000000000000F9 |
| 7480000000000000FF665F720400000000FF0000070804000000FF0000000000000000F9 | 7480000000000000FF665F720400000000FF0000070804000000FF0000000000000000F9 |
| 7480000000000000FF665F720400000000FF0000038404000000FF0000000000000000F9 | 7480000000000000FF665F720400000000FF0000038404000000FF0000000000000000F9 |
| 7480000000000000FF665F720400000000FF00000A8C04000000FF0000000000000000F9 | 7480000000000000FF665F720400000000FF00000A8C04000000FF0000000000000000F9 |
| 7480000000000000FF6600000000000000F8                                     | 7480000000000000FF6600000000000000F8                                     |
| 7480000000000000FF6600000000000000F8                                     | 7480000000000000FF6600000000000000F8                                     |
| 7480000000000000FF665F720400000000FF0000096004000000FF0000000000000000F9 | 7480000000000000FF665F720400000000FF0000096004000000FF0000000000000000F9 |
+--------------------------------------------------------------------------+--------------------------------------------------------------------------+
11 rows in set, 11 warnings (0.023 sec)

4. What is your TiDB version? (Required)

tidb_version(): Release Version: v4.0.0-beta.2-2292-g4218f2836-dirty
Edition: Community
Git Commit Hash: 4218f2836bb38ec79fd080fa88d09d3fe3766c3a
Git Branch: HEAD
UTC Build Time: 2021-03-07 19:53:36
GoVersion: go1.13.4
Race Enabled: false
TiKV Min Version: v3.0.0-60965b006877ca7234adaced7890d7b029ed1306
Check Table Before Drop: false

[kolbe]

5.7.25-TiDB-v4.0.0-beta.2-2292-g4218f2836-dirty (root) [test]> show warnings;
+---------+------+---------------------------------------------------------------------------------------------------------------------------------------+
| Level   | Code | Message                                                                                                                               |
+---------+------+---------------------------------------------------------------------------------------------------------------------------------------+
| Warning | 8045 | invalid record key - "7480000000000000665f720400000000000004b0040000000000" invalid encoded key, expected length: 9, actual length: 6 |
| Warning | 8045 | invalid record key - "7480000000000000665f7204000000000000012c040000000000" invalid encoded key, expected length: 9, actual length: 6 |
| Warning | 8045 | invalid record key - "7480000000000000665f72040000000000000258040000000000" invalid encoded key, expected length: 9, actual length: 6 |
| Warning | 8045 | invalid record key - "7480000000000000665f720400000000000005dc040000000000" invalid encoded key, expected length: 9, actual length: 6 |
| Warning | 8045 | invalid record key - "7480000000000000665f72040000000000000834040000000000" invalid encoded key, expected length: 9, actual length: 6 |
| Warning | 8045 | invalid record key - "7480000000000000665f72040000000000000708040000000000" invalid encoded key, expected length: 9, actual length: 6 |
| Warning | 8045 | invalid record key - "7480000000000000665f72040000000000000384040000000000" invalid encoded key, expected length: 9, actual length: 6 |
| Warning | 8045 | invalid record key - "7480000000000000665f72040000000000000a8c040000000000" invalid encoded key, expected length: 9, actual length: 6 |
| Warning | 1105 | invalid record/index key: 748000000000000066                                                                                          |
| Warning | 1105 | invalid record/index key: 748000000000000066                                                                                          |
| Warning | 8045 | invalid record key - "7480000000000000665f72040000000000000960040000000000" invalid encoded key, expected length: 9, actual length: 6 |
+---------+------+---------------------------------------------------------------------------------------------------------------------------------------+
11 rows in set (0.000 sec)

[tangenta]

This is expected because of the implementation of SPLIT TABLE(on a clustered index table):

The clustered primary key's upper bound & lower bound values are converted to []byte first, then it calculates the cut points without the knowledge of "column data".

tidb/executor/split.go

Lines 256 to 271 in 602bcd2

	func getValuesList(lower, upper []byte, num int, valuesList [][]byte) [][]byte {
	commonPrefixIdx := longestCommonPrefixLen(lower, upper)
	step := getStepValue(lower[commonPrefixIdx:], upper[commonPrefixIdx:], num)
	startV := getUint64FromBytes(lower[commonPrefixIdx:], 0)
	// To get `num` regions, only need to split `num-1` idx keys.
	buf := make([]byte, 8)
	for i := 0; i < num-1; i++ {
	value := make([]byte, 0, commonPrefixIdx+8)
	value = append(value, lower[:commonPrefixIdx]...)
	startV += step
	binary.BigEndian.PutUint64(buf, startV)
	value = append(value, buf...)
	valuesList = append(valuesList, value)
	}
	return valuesList
	}

So the cut points are probably not decodable ;(

[wjhuang2016]

It's the expected behavior and the error should be more understandable.

[tangenta]

@lilinghai Could we remove the critical tag now?

There are two possible solutions for this issue:

• Leave the cut points undecodable. Refine the error message.
• Design a new SPLIT TABLE algorithm for the clustered index tables so that the cut points can be decoded.

WDYT @coocood @bb7133

[bb7133]

Basically I agree with @tangenta , and is there any possibility that we can decode the cut points? If possible, we can turn it into an 'enhancement' or 'feature request'.

I just found that we don't have any user doc for TIDB_DECODE_KEY :(

[coocood]

The issue not only relates to the clustered index, when you split an index, the region's bound key can not be decoded too.