fix: [alert] KubeletTooManyPods (cluster) (#937 )

2026-09-23 19:44:05 +00:00
1 changed files with 6 additions and 0 deletions
@@ -57,6 +57,12 @@ machine:
    extraConfig:
      imageGCHighThresholdPercent: 70
      imageGCLowThresholdPercent: 60
+      # Default 110 leaves no N-1 headroom: with ~207 pods on three nodes, a
+      # rolling drain puts the survivors at 105/110 and KubeletTooManyPods
+      # fires (#937). The ceiling is IP space, not the kubelet: Cilium runs
+      # ipam.mode=kubernetes on the node's /24 podCIDR (~253 usable after
+      # cilium_host), so 200 keeps a margin for pods still terminating.
+      maxPods: 200
 cluster:
  # etcd peers on node0, VLAN 46 (#576). Pinned: Talos never updates an
  # existing member's peer URLs, so changing this subnet means leave + rejoin