Replicating from a newer major version to an older major version in MySQL (for example a 5.6 master and a 5.5 replica) is generally not recommended, but when upgrading a master-master replication topology it’s hard to avoid this scenario entirely. We ended up in this situation last week when upgrading the passive master of an active-passive master-master pair from 5.5 to 5.6. The primary replication flow was going from the active master (5.5) to the passive master (5.6) with no errors, but pt-heartbeat was running on the passive master, which led to a replication failure with this error on the active master:

Last_IO_Error: Got fatal error 1236 from master when reading data from binary log: ‘Slave can not handle replication events with the checksum that master is configured to log; the first event ‘bin-log.002648’ at 4, the last event read from ‘/var/lib/mysqllogs/bin-log.002648’ at 120, the last byte read from ‘/var/lib/mysqllogs/bin-log.002648’ at 120.’

Why did this happen? Starting in MySQL 5.6.6, the new binlog_checksum option defaults to CRC32. Since that option did not exist in MySQL 5.5, the replica can’t handle the checksums coming from the master. Therefore I recommend setting binlog_checksum=NONE in my.cnf as part of the upgrade process for a master-master setup to avoid this error.

My fix was to run this on the passive master:

1

set global binlog_checksum='NONE';

Then I added this to my.cnf so it would survive a restart:

1

binlog_checksum=NONE

After that change was made I examined the binary log to confirm that it did not include anything other than pt-heartbeat activity, and then executed CHANGE MASTER on the active master to skip the checksummed events.

Once the other master is upgraded I can go back and consider changing binlog_checksum to CRC32.