Hi All,

I’m facing a weird error. We’re migrating our old 10.1 multi-master setup to a new 10.3 cluster (everything is running on Debian). New-node01 is connected as a slave to the old cluster’s master.

When I tried to set up the replication between the 2 new nodes (with data files copy), the slave thread stopped nearly immediatelly stating, that the command is invalid – containing illegal characters. After some investigation, it turned out, that binlog on the master (new-node01) is – let’s say – corrupted. After a few iterations/checking I saw, that the error always happens with some MEMORY tables (I know I can exclude them from the replica, but want to understand the issue first).

new-node01/02 running 10.3.27-MariaDB-0+deb10u1-log on latest Debian 10.9 (replication from old-node is OK), replication is GTID based. HW is a new HP Gen10 DL380.

When checking the binlog on the new-node01 with mysqlbinlog, I get the following:

# at 2361

#210407 11:13:34 server id 4  end_log_pos 2403 CRC32 0x90080d27         GTID 100-4-37336878143 trans

/*!100001 SET @@session.gtid_seq_no=37336878143*//*!*/;

START TRANSACTION

/*!*/;

# at 2403

#210407 11:13:34 server id 4  end_log_pos 2527 CRC32 0x27183d2f         Query   thread_id=341297541     exec_time=318   error_code=0

SET TIMESTAMP=1617786814/*!*/;

DELETE FROM `mydb1`.`collect_output`e<FB>#<98>

/*!*/;

# at 2527

#210407 11:13:34 server id 4  end_log_pos 2598 CRC32 0x427f0ed7         Table_map: `mydb1`.`collect_output` mapped to number 1330

Please notice the illegal chars at the end of the „DELETE FROM” line.

(DELETE FROM is normal, since new-node01 is starting up from shutdown and mydb1.collect_output is a MEMORY table -> will delete the contents, so that the replications slaves have the same data).

Is this a bug or a known issue? Is there any other way to solve this, than exclude the MEMORY tables from the replication?

Thanks,

             Tom