Second run got past the install, so the backupstrategy-controller fix was
right, and the capacity preflight reported what this branch exists to find
out:
srv1: requests 2440m / allocatable 3950m = 61%
srv2: requests 2445m / allocatable 3950m = 61%
srv3: requests 2425m / allocatable 3950m = 61%
TOTAL: requests 7310m / allocatable 11850m, worst node 61%
So -smp 4 schedules the lean platform with room to spare, evenly balanced,
and `Insufficient cpu` never appears.
It then failed in the OpenAPI step, at "Test kinds", which reads
.items[0].kind from /apis/apps.cozystack.io/v1alpha1/ingresses and got null
because turning ingress off on the root Tenant means no Ingress instance
exists. That assertion is about the aggregated API returning typed kinds for
list items and has nothing to do with capacity, so the fix is to give it an
instance rather than to weaken it: ingress goes back on, and the HR waits name
it again.
It costs about 310m for the NGINX controller and its default backend, against
39 points of headroom, so the experiment is unaffected. monitoring and
seaweedfs stay off, which is where the bulk of the tenant-root workload was.
Checked the rest of that file for the same shape: "Test kinds" is the only
place an app instance is required, and the only two kinds it reads are
tenants and ingresses. Nothing in it references monitoring, seaweedfs, bucket,
grafana, velero, goldpinger or backupstrategy.
Assisted-By: Claude <noreply@anthropic.com>
Signed-off-by: Myasnikov Daniil <myasnikovdaniil2001@gmail.com>
Myasnikov Daniil
9f8d6123fd