|Traditional replication and mysql-galera/wsrep||Kian||4/23/12 10:33 AM|
I currently have a 3-node Galera cluster. The first node is receiving
replication data from a separate (non-galera) MySQL server, using
traditional MySQL replication. No other writes are happening because
this is a test cluster.
On node1, I have wsrep_cluster_address="gcomm://". On node2 and
node3, I have the cluster address set to the address of node1.
Over the weekend, it appears MySQL on node1 crashed (out of memory)
but mysqld_safe restarted it. When I checked this morning, I noticed
a wsrep_cluster_size of '1' on node1, and wsrep_cluster_size of '2' on
node2 and node3. In other words, node1 appeared to be disconnected
from the cluster.
However, replication still seemed to be working (IO/SQL threads
running, 0 seconds behind master).
1) This sounds like a split-brain problem, because I don't believe any
of the writes (via replication) could have been propagated to node2 or
node3, but they still succeeded on node1. How could this happen, with
3 nodes in the cluster?
2) When this happens, what should I do? Should I set
wsrep_cluster_address on node1 to the address of either node2/node3?
In this case, I will be losing data, correct?
3) After initially setting up a cluster, should I change the
wsrep_cluster_address of the primary node, instead of leaving it as an
Thanks for all of your help.
|Re: [codership-team] Traditional replication and mysql-galera/wsrep||Alexey Yurchenko||4/23/12 11:14 AM|
This is not split brain problem, this is a new cluster problem.
> 2) When this happens, what should I do? Should I set
In this particular case you should reconnect nodes 2/3 to new cluster
> 3) After initially setting up a cluster, should I change the
You should do exactly that.
> Thanks for all of your help.