https://rancher.com/ logo
Title
f

fancy-insurance-98888

04/07/2023, 5:15 AM
Hi, I deployed a HA-cluster with two master and a mysql instance. I noticed that there has a huge throughout from mysql to one of masters as shown in the page below all the time. Is this normal or how can I debug it? Thanks.
c

creamy-pencil-82913

04/07/2023, 7:24 AM
its pretty chatty, yeah. The throughput will be proportional to the load on the apiserver.
s

sparse-fireman-14239

04/07/2023, 8:29 AM
For a HA setup you need 3 or 5 control plane nodes.
c

creamy-pencil-82913

04/07/2023, 8:57 AM
not with external sql you don’t
f

fancy-insurance-98888

04/07/2023, 9:58 AM
@creamy-pencil-82913 You mean this is normal,right? Meanwhile, another master has little throughout, no load-balance. @sparse-fireman-14239 In my opinion, the apiserver is stateless, several apiservers just for load-balance, not for election,right?Additionally,two masters are written in docs(Two or more server nodes) :)
r

rough-farmer-49135

04/07/2023, 10:43 AM
Well...technically you're HA as long as the external database stays up. Is your MySQL instance running on >1 node?
c

creamy-pencil-82913

04/07/2023, 2:34 PM
Correct, apiservers are stateless but the controllers are leader-elected and talk to the local apiserver. So you will effectively always have one node that's active and others standby
The odd number of nodes restriction is just for etcd and doesn't apply if you're using an external datastore
f

fancy-insurance-98888

04/10/2023, 1:51 AM
@rough-farmer-49135 Yes, there is only one MySQL instance running on one node.
r

rough-farmer-49135

04/10/2023, 1:26 PM
That means that if you want to test how things work with an HA Kubernetes cluster, you're good. However if you're trying to protect yourself from a single point of failure, you've just moved it back from the control plane to the MySQL instance.