Dajka Tamás <viper@vipernet.hu> writes:
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>
Looks very much related to binlog checksum. Cannot say anthing more, maybe you could develop it further knowing details that unreported here? Cheers, Andrei
/*!*/;
# 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.