Open capgadsx opened 7 years ago
Maybe this is a good start. https://github.com/kubernetes/kubeadm/issues/277
We were thinking with @capgadsx that it will be a good solution to put kubemaster and etcd in a VM in Ovirt and do backups in lustre with rsync, so the HA is up to Ovirt. With a good VM configuration(auto net, auto service up in restarts, etc) the only enemy will be a IML server's failure or an entire datacenter failure.
That is overthink. 3 etcd is enough.
I agree with @sirspock. Please do not complicate things more than it will be needed. If 3 independent disks of 3 different servers fail at the same time, everyone will understand a 503!
@capgadsx can you edit your first comment to add a brief doc of the 3 etcd solution we are currently using? (just for the sake of documentation)
At this point we already implemented the 3-member etcd cluster and we have 3 kubernetes api-servers running. We still have to setup the load-balancer for the api-servers
The idea is to implement the solution described in the official kubernetes documentation, using a clustered etcd deployment and load-balanced api-servers.