executor: replace should not change other rows when auto ID is out of range

[tangenta]

What problem does this PR solve?

Issue Number: close #29483

Problem Summary:

SET sql_mode="NO_ENGINE_SUBSTITUTION";
DROP TABLE IF EXISTS t1;
CREATE TABLE t1 (a tinyint not null auto_increment primary key, b char(20));
INSERT INTO t1 VALUES (127,"maxvalue");
REPLACE INTO t1 VALUES (0,"newmaxvalue");

Specifying a zero value for the auto_increment column triggers ID allocation. In the current implementation, the allocated ID will be replaced in the row which is equivalent to

REPLACE INTO t1 VALUES (128,"newmaxvalue");

Since 128 is out of range for tinyint column, 128 is truncated to 127 in setDatumAutoIDAndCast():

REPLACE INTO t1 VALUES (127,"newmaxvalue");

However, these two REPLACE statements should be different because

• for the 0 case, it means adding a new row.
• for the 127 case, it means updating an old row if exists.

What is changed and how it works?

In this PR, I add a check inside setDatumAutoIDAndCast() to prevent a truncated auto-ID is used. The error message is different from MySQL's "duplicate primary key" because of the implementation difficulty: there is no way to tell ReplaceExec that whether the row is filled by auto-IDs without changing the interface.

tidb/executor/replace.go

Line 192 in a046014

func (e *ReplaceExec) exec(ctx context.Context, newRows [][]types.Datum) error {

This can cause MySQL incompatibility when there is no duplicated auto-ID:

SET sql_mode="";
CREATE TABLE t1 (a tinyint not null auto_increment primary key, b char(20));
INSERT INTO t1 VALUES (127,"maxvalue");
DELETE FROM t1;
REPLACE INTO t1 VALUES (0,"newmaxvalue"); -- MySQL executes successfully
-- ERROR 1467 (HY000): Failed to read auto-increment value from storage engine

I think it is fine because this error can be an alert, reminding the users that auto-IDs are exhausted.

Check List

Tests

• Unit test
• Integration test
• Manual test (add detailed scripts or steps below)
• No code

Side effects

• Performance regression: Consumes more CPU
• Performance regression: Consumes more Memory
• Breaking backward compatibility

Documentation

• Affects user behaviors
• Contains syntax changes
• Contains variable changes
• Contains experimental features
• Changes MySQL compatibility

Release note

Fix an issue that REPLACE statement changing other rows when the auto ID is out of range

[ti-chi-bot]

[REVIEW NOTIFICATION]

This pull request has been approved by:

• bb7133
• wjhuang2016

To complete the pull request process, please ask the reviewers in the list to review by filling /cc @reviewer in the comment.
After your PR has acquired the required number of LGTMs, you can assign this pull request to the committer in the list by filling /assign @committer in the comment to help you merge this pull request.

The full list of commands accepted by this bot can be found here.

Reviewer can indicate their review by submitting an approval review.
Reviewer can cancel approval by submitting a request changes review.

[bb7133]

LGTM, the behavior is different with MySQL but I think it is reasonable to reports an error when auto_inc capacity exceeded.

Actually, I got surprised that MySQL gives a 'duplicates error' for REPLACE. /cc @morgo @Alkaagr81

[bb7133]

/merge

[ti-chi-bot]

This pull request has been accepted and is ready to merge.

Commit hash: 31f7e39

[ti-srebot]

cherry pick to release-5.3 in PR #32324

[guo-shaoge]

@tangenta could you please work on this cherry pick to v5.3. Thank!

[tangenta]

/run-cherry-picker

[ti-srebot]

cherry pick to release-5.2 in PR #33863