Akash - nodes tainted as not compute still get counted

[avolon42x]

I ran into a problem where my config has some nodes with the flag "is compute node" not activated. The available CPU still gets counted from theses nodes and thus I ended up having some leases not able to run since there was no more CPU left. So kubernetes still reports all available nodes no matter if they are "compute node" or not.

[alexsmith540]

i'll have to dig into this. Last time I tested moving a compute node to not-compute it worked ok but would have to re-test.
Edit: I re-read this just now, duh. Are there differences you see in the provider logs stated CPU values versus what you actually have available? I'll have to check as well if the CPU values in the handyhost UI (in the donut charts) also reflect non-compute nodes.

[avolon42x]

Yeah I wasn't precise enough:
So I have now my cluster setup so that the master is not part of the compute nodes. All the following numbers include the master though:
Dashboard reports: 1112m used and 49000m total (capacity)
Allocatable according to the dashboard: 48300m
Logs report: "15325", "15395", "10625", "1625", "880", "1525" = 45375m

I understand that there are variances since Kubernetes is also running etc.
If I rebuild the cluster with let's say etcd not applied as compute node the cluster is always shown with an error (yellow - node not available). It only works with the master node not part of the compute nodes - but the capacity is still made available which leads to overallocation..