We are using 6 nodes cluster. 3 nodes located in NYC and 3 nodes located in California.
We are using Mysql for databases and installed Redhat SSO on redhat 8. Every time we have a network or power issue in one datacenter the cluster crashes and and application is not available.
Our expectation was that the application would fail over to available datacenter.
- How to fix the HA to work properly so that when we have and outage the application is accessible in the available datacenter?
- How do a make this a true HA ?
Hello @tdiallo ,
This is a bad design. You will have split-brain if the network goes down.
Yes, see above.
Have an odd number of nodes total. You can create a garbd process in a 3rd datacenter which will function as a additional voting member. If the link between dc1 and dc2 goes down, as long as dc1->dc3 and dc2->dc3 remains, then the cluster will remain online.
If dc1 goes completely offline, dc2 will remain as it has 3 nodes + the arbitrator (4 nodes) which is > 50% of the total number of nodes (7).
If dc2 goes completely offline, dc1 will remain as it has 3 nodes + the arbitrator (4 nodes) which is > 50% of the total number of nodes (7).
Thank you … this is very informative