Keep your EKS Kubernetes events for months, and find what will block your next node upgrade
Kubernetes deletes events after one hour. When a pod crash-looped at 3 AM or a node group upgrade failed with
PodEvictionFailure, the evidence is usually gone by the time you look. Connect an AWS account with a narrow
read-only role and an EKS access entry: we read the cluster's events every few minutes from outside the cluster, keep
them for up to 90 days, and let you search, export and alert on them. Nothing is installed in your cluster.
Event history
- Search by cluster, namespace, kind, object name, reason, Warning or Normal, text and time range; download CSV
- Repeats of the same event are kept as one row with a count, first seen and last seen
- Alert rules to Slack or email, for example crash loops, failed scheduling, OOM kills and evictions, or any Warning in a namespace
- One alert per object and reason per cooldown, silence windows, and no burst of old events when you connect
- A clear note when events may be missing because the cluster could not be read
Upgrade blocker check
- PodDisruptionBudget allows 0 disruptionsScale the workload above the budget (or fix its unhealthy pods), or change the budget to maxUnavailable: 1, then retry the upgrade.
- Single replica protected by a PodDisruptionBudgetRun at least 2 replicas, or use maxUnavailable: 1 for this budget. With 1 replica the only pod can never be evicted.
- Pod tolerates every taintGive the toleration a key (for example only node.kubernetes.io/not-ready). Tolerating everything lets the pod land back on the node being drained.
- Pod without a controllerRun it from a Deployment, StatefulSet or Job. A drain deletes a bare pod and nothing recreates it, so EKS stops unless you force the update.
- Node group update failed: PodEvictionFailureFix the pods named here (usually their PodDisruptionBudget), then start the update again. Force update evicts them anyway, with downtime.
- Node group health issueResolve the issue AWS reports for the node group before upgrading it.
Runs daily and on demand, and names the exact pods and PodDisruptionBudgets.
Read-only: the IAM role can only list and describe EKS clusters, node groups and updates. Inside the cluster it has the AWS-managed view policy (no Secrets). Limitation: clusters whose API endpoint is private only cannot be reached.