Skip to content

docs: no guidance on raising max-pods for large nodes #64

Description

--kubelet-arg=max-pods=220 is in all three example inventories and in the README table, with "Cozystack runs many pods per node" as the reason. There is nothing about when to raise it or how to pick a value, so it reads like a fixed requirement.

On a 128 CPU / 1.5 TiB node this turned into the binding constraint: the node sat at 220 pods with 6% of memory used, and platform pods stayed Pending. The only clue is in the scheduler events, 0/1 nodes are available: 1 Too many pods, which is easy to miss when the node looks empty in every other metric.

Raising it needs one number and one restart, but the value depends on things the README does not mention. Pod IP capacity is the real ceiling and it is not the per-node podCIDR: with kube-ovn the addresses come from the shared ovn-default subnet, so a /16 there means the /24 podCIDR is irrelevant.

Would be useful to have a short note next to that table: what 220 is sized for, what to check before raising it (subnet capacity, kubelet guidance), and one example inventory line with a larger value.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions