The testcase was derived from

   https://bugs.mysql.com/bug.php?id=88321

and especially

   https://bugs.mysql.com/file.php?id=26087&bug_id=88321&text=1

I changed a lot within the mechanics in order to have finally some

handy test using MTR + the mysql client.

Scenario

-------------

The table definition is

CREATE TABLE test (

   test_id int(11) NOT NULL AUTO_INCREMENT,

   testcol varchar(256) COLLATE utf8_bin NOT NULL,

   PRIMARY KEY (test_id) )

ENGINE=InnoDB DEFAULT CHARSET=utf8 COLLATE=utf8_bin;

Two "worker" run concurrent a big number of loops containing

    INSERT INTO test (testcol) VALUES ('test123');

    .... Initiate the abort of the test in case that fails

    # The RELEASE is essential.

    COMMIT WORK RELEASE;

   # Dummy statement for detecting the loss of the connection

   COMMIT ;

   ... get a reconnect ...

One "mad DBA" runs concurrent to the two "worker" above a big number of loops containing

    SET @@GLOBAL.auto_increment_increment = 1;

    sleep 0.1;

    SET @@GLOBAL.auto_increment_increment = 2;

    sleep 0.1;

After around ~ 20s one of the two sessions initiates the abort because it

harvested something like a

ERROR 23000: Duplicate entry '331' for key 'PRIMARY'

10.1 commit 403e6615fdd3f5f4de6d159457b2bd2fe566e0ad 2019-07-26

bb-10.1-release commit 4177181e1617df185397ea275ee98eb4d50a0b8b 2019-07-26

    # It was told that bb-10.1-release might contain a fix which fixes one or

    # more weaknesses around AUTO_INCREMENT value generation.

10.2 / 10.3 / 10.4 2019-07