Date: 2026-07-21
Status: Applied
Node: drg-k8s.auth.home.lan
Raised fs.inotify.max_user_instances from 128 to 1024 and persisted it in:
/etc/sysctl.d/99-kubernetes-inotify.conf
Configuration:
# Kubernetes and containerd run many root-owned filesystem watchers.
fs.inotify.max_user_instances = 1024
Restarted containerd after applying the setting. Kubernetes reconciled the node and workloads without an ingress deployment mutation.
Containerd's CRI plugin could not initialize its CNI filesystem watcher because the per-user inotify instance limit was exhausted. This made kubelet lose CRI access and caused cluster ingress to stop serving application traffic.
sysctl -n fs.inotify.max_user_instances: 1024Related incident: Containerd CRI unavailable after inotify exhaustion