Details
-
Bug
-
Status: Open (View Workflow)
-
Major
-
Resolution: Unresolved
-
11.4.2
-
None
-
None
Description
Since we recently upgraded all our MariaDB/Galera clusters from 10.11.7 to 11.4.2, we noticed that the second part of the GTID for Galera set-ups changed.
As per documentation we choose to set a different server_id on each cluster node. This way we can track each transaction down to the originating node in our binlog analysis tooling.
https://mariadb.com/kb/en/using-mariadb-replication-with-mariadb-galera-cluster-using-mariadb-replica/#setting-server_id-on-cluster-nodes
https://mariadb.com/kb/en/gtid/#implementation
Relevant config node1
gtid_domain_id = 10168
|
server_id = 10168
|
wsrep_gtid_domain_id = 10160
|
Relevant config node2
gtid_domain_id = 10169
|
server_id = 10169
|
wsrep_gtid_domain_id = 10160
|
Relevant config node3
gtid_domain_id = 10170
|
server_id = 10170
|
wsrep_gtid_domain_id = 10160
|
DML on node2
UPDATE t1 SET b = 'b' WHERE i = 1;
|
Binlog on all nodes
#240705 12:00:38 server id 10168 end_log_pos 1314 CRC32 0xb816275b GTID 10160-10168-4015672 trans
|
As you can see, the second part (server_id) corresponds with node1 and not with the originating node2.
I'm not sure which server-id is used, possible that from the first (bootstrapped) node.
The change doesn't break replication (since domain-id and seqnr is the unique part), but is does break the tracking of originating node.