How worker to worker nodes communication works in Kamaji? #406
Replies: 1 comment 2 replies
-
The communication between the Worker nodes and the Control Plane is definitely the same as you would have on a traditional Kubernetes cluster: the Control Plane endpoint refers to the IP used by your Tenant Control Plane. This is trivial in flat networks, things get more complicated if the Control Plane is deployed in a private subnet, different from the worker nodes ones: and there are many reasons for this kind of setup, mostly for security reasons. In a such case, the worker nodes wouldn't be routable from the Kubernetes API Server, and here comes the About the last question, there are no big issues with latency between the CP and WN although some considerations must be put in place:
The design you suggested is ok from a theoretical point of view, although this implies some variables out of your control, such as the WAN, such as network failures, or the physical layer issues (sharks love to crunch cables). As I always suggest, it's possible, you just need to evaluate the pros and cons. |
Beta Was this translation helpful? Give feedback.
-
Hello. I'm still new to Kubernetes, and I don't understand how worker nodes communicate with other worker nodes in Kamji. Do they communicate through the Konnectivity agent on each worker node without going through the control planes first? Also, I want to ask if it's okay to have a central Tenant Control Plane in one region, for example, Vultr Singapore, and have worker nodes in another region, such as Hetzner Finland?
Beta Was this translation helpful? Give feedback.
All reactions