[Maria-discuss] Async replication between two Galera clusters, GTIDs on non slave node in "slave cluster" don't match
Hi! I have two Galera clusters of three nodes each, let's call them cluster A and cluster B. Cluster B has one member set up as a slave to one of the members in cluster A. The cluster B member that runs the slave protocol shows it's GTID matches that of the nodes in cluster A, but the other two nodes in cluster B (not running slaves themselves, but in the cluster with the node running as a slave) have GTIDs that are always 4 behind the other nodes. So just to lay it out again: A0 5-24-5 - master A1 5-24-5 A2 5-24-5 B0 5-24-5 - slave to A0 B1 5-24-1 B2 5-24-1 The GTIDS stay 4 apart, and replication seems fine. I add rows on any A cluster node and they show up in all three B nodes. The GTID on the node running slave stays in sync with the A cluster. The A0 cluster is set up using xtrabackupv2 for SST. The A cluster is completely up before doing anything on the B cluster. The B cluster is started like: 1. Start B0 as first node in the Galera cluster 2. mysqldump from A0 to B0 with --gtid. <https://gist.github.com/warnerpr/011cabb876600c33e5bee42d7d0b17d2> https://gist.github.com/warnerpr/011cabb876600c33e5bee42d7d0b17d2 3. Master of B0 is set to A0 4. Slave is started on B0 5. B1 and B2 are started and set to join the existing cluster. Cluster A and B have log_slave_updates = ON. Details: B0 Slave - https://gist.github.com/warnerpr/59c287bb201302961742bb89d5035559 B1 - https://gist.github.com/warnerpr/0fb8825e49bce72761e0bdd95b0512d9 A1 Master - https://gist.github.com/warnerpr/91e0df1eb04f425baca3675335639c0f Any pointers are greatly appreciated! Paul
Hi Paul, I don't think that will work across clusters. Your best bet is usually not to rely on GTID for cross-cluster replication. Regards GL Le jeu. 2 nov. 2017 à 06:30, Warner, Paul <pwarner@ciena.com> a écrit :
Hi!
I have two Galera clusters of three nodes each, let's call them cluster A and cluster B.
Cluster B has one member set up as a slave to one of the members in cluster A.
The cluster B member that runs the slave protocol shows it's GTID matches that of the nodes in cluster A, but the other two nodes in cluster B (not running slaves themselves, but in the cluster with the node running as a slave) have GTIDs that are always 4 behind the other nodes.
So just to lay it out again:
A0 5-24-5 - master
A1 5-24-5
A2 5-24-5
B0 5-24-5 - slave to A0
B1 5-24-1
B2 5-24-1
The GTIDS stay 4 apart, and replication seems fine. I add rows on any A cluster node and they show up in all three B nodes. The GTID on the node running slave stays in sync with the A cluster.
The A0 cluster is set up using xtrabackupv2 for SST.
The A cluster is completely up before doing anything on the B cluster.
The B cluster is started like:
1. Start B0 as first node in the Galera cluster 2. mysqldump from A0 to B0 with --gtid. <https://gist.github.com/warnerpr/011cabb876600c33e5bee42d7d0b17d2> https://gist.github.com/warnerpr/011cabb876600c33e5bee42d7d0b17d2 3. Master of B0 is set to A0 4. Slave is started on B0 5. B1 and B2 are started and set to join the existing cluster.
Cluster A and B have log_slave_updates = ON.
Details: B0 Slave - https://gist.github.com/warnerpr/59c287bb201302961742bb89d5035559 B1 - https://gist.github.com/warnerpr/0fb8825e49bce72761e0bdd95b0512d9 A1 Master - https://gist.github.com/warnerpr/91e0df1eb04f425baca3675335639c0f
Any pointers are greatly appreciated!
Paul
_______________________________________________ Mailing list: https://launchpad.net/~maria-discuss Post to : maria-discuss@lists.launchpad.net Unsubscribe : https://launchpad.net/~maria-discuss More help : https://help.launchpad.net/ListHelp
participants (2)
-
Guillaume Lefranc -
Warner, Paul