Skip to main content

Container Restart Check

Last updated on

Container Restart Check is a built-in Command Probe template that validates whether the restart count of a container stays within an acceptable threshold during a chaos experiment. Use it to confirm that a workload does not enter a restart or crash loop while a fault stresses the application. You select pods by label, by name, or by the owning workload kind and namespace.

The probe runs the healthchecks utility bundled in the chaos probe image, queries the Kubernetes API, and prints [Pass] when the restart count of the targeted container is at or below CONTAINER_RESTART. The comparator marks the probe as passed when the output contains [Pass].

Built-in probe template

This is a built-in Command Probe template that runs on Kubernetes chaos infrastructure. Add it to an experiment from the probe library and customize its inputs. Go to Built-in probe templates to browse the full library, or go to Command probe to understand how command probes work.


Use cases

Use this probe template to:

  • Verify that containers do not restart excessively during chaos experiments.
  • Monitor container stability during resource stress.
  • Validate application resilience to failures.
  • Confirm that pods maintain healthy restart counts.

How the probe works

The template configures a Command Probe that runs healthchecks -name validate-container-restart. The utility resolves the target pods from TARGET_LABELS, TARGET_NAMES, TARGET_KIND, and TARGET_NAMESPACE, reads the restart count of the container named in TARGET_CONTAINER, and prints [Pass] when the count is at or below CONTAINER_RESTART. The comparator passes the probe when the output contains [Pass], and fails it otherwise.


Prerequisites

  • Chaos infrastructure: A Kubernetes chaos infrastructure installed in the target cluster.
  • Namespace access: Access to the target namespace and pods.
  • RBAC permissions: Permissions for the chaos service account to query pod status.

Probe properties

Command

healthchecks -name validate-container-restart

Comparator

TypeCriteriaValue
stringcontains[Pass]

The probe passes when the command output contains [Pass], which indicates that container restart counts are at or below the acceptable threshold.

Environment variables

VariableDescriptionRequiredDefault
TARGET_LABELSComma-separated list of labels used to filter pods (for example, app=nginx,env=prod).No-
TARGET_NAMESComma-separated list of target pod names.No-
TARGET_NAMESPACENamespace of the target pods.Yes-
TARGET_KINDKind of the owning workload (for example, deployment, statefulset, daemonset).Nodeployment
TARGET_CONTAINERName of the container to check the restart count for.No-
CONTAINER_RESTARTMaximum allowed restart count. The restart count must be equal to or less than this value.No1
STATUS_CHECK_TIMEOUTMaximum time in seconds to wait for the status check.No180
STATUS_CHECK_DELAYDelay in seconds between status checks.No2

Run properties

PropertyDescriptionTypeDefault
timeoutMaximum time to wait for the probe to complete (for example, 30s, 1m, 5m).String180s
intervalTime between probe executions (for example, 1s, 5s, 10s).String1s
attemptNumber of retry attempts before the probe is marked as failed.Integer1
pollingIntervalTime between retry attempts (for example, 1s, 5s, 10s).String-
initialDelayInitial delay before the probe starts (for example, 0s, 10s, 30s).String-
stopOnFailureStop the experiment if the probe fails.Booleanfalse
verbosityLog verbosity level (info, debug, trace).String-

Troubleshooting

Container Restart Check probe fails because the container exceeded the restart threshold

The container restart count is higher than CONTAINER_RESTART, which usually means the workload is crash-looping under the injected fault. Inspect the pod with kubectl describe pod and check container logs with kubectl logs --previous to find the crash cause, then either fix the workload or raise CONTAINER_RESTART if a higher count is acceptable.

Container Restart Check probe fails because no pods or containers matched

The selectors did not resolve any pods, or TARGET_CONTAINER did not match a container in the resolved pods. Confirm that TARGET_LABELS, TARGET_NAMES, TARGET_NAMESPACE, and TARGET_KIND identify the workload, and that TARGET_CONTAINER matches a container name in the pod spec.

Container Restart Check probe fails with a forbidden or RBAC error

The chaos service account does not have permission to read pods in the target namespace. Grant get and list on pods for the chaos service account in that namespace, then rerun the experiment.