In your example, the behavior is expected :
ORM code

		for (var demo : demoList) {

			if (i > 0 && i % BATCH_SIZE == 0) {

				entityManager.flush();

				entityManager.clear();

			}

			entityManager.persist(demo);

			i++;

		}

Will result in hibernate internally executing 5 * JDBC prepareStatement.execute().
Since there is 2 paralleles runner, this result with INSERT with mixed IDs, like you describe.
There is nothing wrong in MariaDB Server or connector there.
This is just how hibernate send the data.

There is some solutions hibernate side, like for example adding batch_size :

spring:

  jpa:

    hibernate:

      jdbc:

        batch_size: 2000

In that case, hibernate will not run some PrepareStatement.execute, but will execute those in batch (preparestatement.addBatch() x times, then preparestatement.executeBatch)
MariaDB connector and Server will then receive only one command with all the insert at a time, then IDs will be orderered.

If you debug the demo, you'll see it's not the way you describe because you seem to have missed this line which sets the bach size only for the current hibernate Session instead of setting it globally like you were suggesting:

entityManager.unwrap(Session.class).setJdbcBatchSize(BATCH_SIZE);

In fact, this part is not executed in the current settings of the demo (4 objects by thread) but I left it to support lists which have a size higher than BATCH_SIZE (50):

entityManager.flush();

entityManager.clear();

I added the following to show what commands the MariaDB connector client was sending:

logging:

  level:

    '[org.mariadb]': DEBUG

And aside from transactions related commands, only two commands are issued by each thread : one prepare (PreparePacket) and one bulk insert (BulkExecutePacket) with no mixed up data when I inspect the batchParameterList field.

sorry, i've made bad assumption. This is indeed an issue connector side : calendar use in prepareStatement.setTimestamp(int parameterIndex, Timestamp x, Calendar cal) is not used in synchronized block, resulting in this issue when use in parallel

problem is only in 3.0 version

corrected in 3.0.9-SNAPSHOT version with commit https://github.com/mariadb-corporation/mariadb-connector-j/commit/22388e698edb045ea9e8c560195caa9bf4bb6da3

available using snapshot repository :

<repositories>

    <repository>

        <id>sonatype-nexus-snapshots</id>

        <name>Sonatype Nexus Snapshots</name>

        <url>https://oss.sonatype.org/content/repositories/snapshots</url>

    </repository>

</repositories>

<dependencies>

    <dependency>

        <groupId>org.mariadb.jdbc</groupId>

        <artifactId>mariadb-java-client</artifactId>

        <version>3.0.9-SNAPSHOT</version>

    </dependency>

</dependencies>

It works great with the demo and also with a real world application, thanks!