Comment 30 for bug 1080765

We've been seeing this issue on our production server with a MariaDB master / MariaDB slave which has really low replication lag. We've been running pt-table-checksum nightly on a cron job, and have recently been seeing false positives.

I found that the initial run was giving false positives on one or two tables, but running --replicate-check-only afterwards does not report any false positives.

I'm guessing that as per previous comments, the slave is slightly behind even though it reports not being behind at all, and is mid-way-though execution of the UPDATE query of the master_* columns, so the SELECT statement to check the difference is checking wrong values.

Adding the argument --replicate-check-retries=2 causes pt-table-checksum to check the checksum a second time if it finds a difference.

In my testing this has eliminated the false positives.