Missing telemetry is the worst kind of bug. Nothing errors. The pods run. The dashboard just shows... nothing.
The Middleware Kubernetes agent installs as a DaemonSet, and the install has four sharp edges. Heres the checklist we keep for it.
# how many copies are running? exactly one per cluster, no more
kubectl get daemonset -A | grep -i middleware
# pods crash-looping? check the reason before restarting them
kubectl get pods -A | grep -i middleware
kubectl logs -l app=middleware-agent --all-containers
1. Two DaemonSets in one cluster. Run exactly one agent DaemonSet per cluster. A second install (a duplicate helm release, an overeager deploy script) fights the first, and data quietly goes missing. Delete the duplicate.
2. Clusters sharing a kube context get the same name. If you manage several clusters from one context, set a distinct cluster name on each install. Without it, every clusters data lands under one name, and some clusters look like they have no data at all.
3. CrashLoopBackOff with OOMKilled. That is the memory limit being too low, not a flaky pod. Raise the limit in the DaemonSet manifest. Restarting and hoping wont fix it.
4. Healthy pods, missing data. Almost always the cluster name or the API key. And keep the key in a Kubernetes Secret, not inline in the manifest.
Full write-up, fetchable by agents via curl:
https://vectle.com/skills/skl_1JJRlHOJrb68wIYl6KWoHQ
(Source docs: https://docs.middleware.io/agent-installation/kubernetes/kubernetes-agent)
Top comments (0)