mirror of
https://github.com/cozystack/cozystack.git
synced 2026-11-03 18:47:04 +09:00
Aleksei Sviridkin
0cb44dc919
The inventory has to run next to kubelet: a DaemonSet reading the host's sysfs and kubelet's pod-resources socket, both mounted read-only. Connecting to the socket needs root, so the agent runs as root but with no capabilities and no privilege escalation. The manager's cached client would watch every pod in the cluster, so the agent uses a direct client and lists only its own node's pods. RBAC cannot name the caller's node, so the role is cluster-wide; it holds only the verbs the agent uses, and the DaemonSet selects the nodes node-feature-discovery marks as carrying an NVIDIA device, so the token does not sit on nodes with nothing to report. The image is in the root build list so that every build path stamps its reference; the committed tag is replaced by the first build. Assisted-by: LLM Signed-off-by: Aleksei Sviridkin <f@lex.la>
16 lines
522 B
YAML
16 lines
522 B
YAML
image: ghcr.io/cozystack/cozystack/gpu-node-agent:latest
|
|
# Go duration between inventory syncs, such as 30s or 1m. kubelet's
|
|
# pod-resources API has no watch.
|
|
interval: 30s
|
|
# The label gpu-operator's node-feature-discovery puts on nodes with an
|
|
# NVIDIA PCI device: elsewhere the agent has nothing to report and its
|
|
# token, which may list every pod, has no business being.
|
|
nodeSelector:
|
|
feature.node.kubernetes.io/pci-10de.present: "true"
|
|
resources:
|
|
requests:
|
|
cpu: 10m
|
|
memory: 32Mi
|
|
limits:
|
|
memory: 64Mi
|