Skip to main content

Pod JVM Kafka latency

Last updated on

Pod JVM Kafka latency is a Kubernetes pod-level chaos fault that adds a configurable delay to Kafka client calls from a JVM running in a target container, scoped to a chosen topic and client mode, for a configurable duration. Only matched calls are slowed; other Kafka traffic and other code paths run at normal speed. When the fault ends, Kafka calls return to baseline latency immediately.

Use this fault to test how a Java service behaves when Kafka becomes slow on a specific code path: a partition leader that takes longer to ack produces, a consumer side processing messages slowly, or a degraded cluster on one topic.

Run your first experiment

If you have not configured the chaos infrastructure yet, go to Quickstart to install the chaos infrastructure and run an experiment end to end.

Use cases​

Run this fault when you want to answer concrete questions like:

  • Producer timeout sensitivity: When send().get() takes 2 seconds instead of 50 ms, does the application respect request.timeout.ms and shed load, or does it block?
  • Consumer lag growth: Slow consumer-side processing and verify whether the application catches up after the fault or rebalances.
  • In-flight buffer pressure: Slow producer ack and see whether buffer.memory fills up and triggers application-level back-pressure.
  • End-to-end SLO impact: Does the added Kafka latency push request p99 over your SLO, or does the design absorb it?
  • Circuit breaker behavior: Does a circuit breaker around the slow operation trip and recover correctly when the fault ends?

Prerequisites​

  • Kubernetes version: 1.21 or later. Go to What's supported to confirm distribution support.
  • Target pod is Running: The Java application pod is in the Running state.
  • Java agent attach available: The Java process allows agent attach. Utilities such as ps, pgrep, and bash are present in the container, and the JVM is not built with a restricted runtime that strips attach modules.
  • Kafka client in classpath: The target JVM uses the Apache Kafka Java client and produces to or consumes from the configured KAFKA_TOPIC in the configured KAFKA_MODE.
  • Privileged pods allowed: The cluster lets you schedule privileged pods in the chaos namespace. GKE Autopilot supports this fault but requires the one-time setup in Chaos on GKE Autopilot; other locked-down distributions may need similar exemptions.
  • Container runtime access: The chaos pod can reach the container runtime socket on the target node (/run/containerd/containerd.sock, /var/run/docker.sock, or /var/run/crio/crio.sock).
  • Workload selector defined: The chaos experiment knows the target workload by kind, namespace, and either names or labels.
JVM chaos uses the Byteman agent

This fault attaches a Byteman agent to the target JVM over BYTEMAN_PORT. The port must be reachable from the chaos pod and must not be in use by the application.

Supported environments​

PlatformSupport status
Amazon EKSSupported
Azure AKSSupported
Google GKESupported
Red Hat OpenShiftSupported
RancherSupported
VMware TanzuSupported
Self-managed Kubernetes (CNCF-certified)Supported
GKE AutopilotSupported with Autopilot setup
EKS Fargate, ACI virtual nodesNot supported (no access to container runtime sockets)

Permissions required​

The fault runs under the chaos infrastructure's service account.

Resource (apiGroup)VerbsWhy it is needed
pods ("")get, list, create, delete, deletecollection, patch, updateDiscover target pods and run the chaos pod on the same node
pods/log ("")get, list, watchStream chaos pod logs for status and debugging
deployments, statefulsets, replicasets, daemonsets (apps)get, listResolve the target workload to the pods it owns
events ("")get, list, create, patch, updateRecord fault progress as Kubernetes events
jobs (batch)get, list, create, delete, deletecollectionRun the chaos job that drives the fault

The default Harness chaos infrastructure service account already includes these permissions.

Fault tunables​

Configure the following fault parameters when you add Pod JVM Kafka latency to an experiment in Chaos Studio. Defaults are shown for reference.

Kafka filters

TunableDescriptionDefault
KAFKA_MODEClient mode to target. One of producer (publish-side) or consumer (subscribe-side)."producer"
KAFKA_TOPICTarget Kafka topic name. Empty matches all topics for the configured mode.""
TRANSACTION_PERCENTAGEPercentage of matched Kafka operations to delay, between 0 and 100. 0 delays none; 100 delays every match.0

Chaos parameters

TunableDescriptionDefault
LATENCYDelay to add to each matched operation, in milliseconds.2000
TOTAL_CHAOS_DURATIONDuration of the fault in seconds.60

JVM

TunableDescriptionDefault
BYTEMAN_PORTPort on which the Byteman agent listens inside the container. Must not conflict with any port already in use.9091
JAVA_HOMEAbsolute path to the Java installation inside the container. Empty auto-detects from PATH.""

Targeting

TunableDescriptionDefault
TARGET_PODSComma-separated list of pod names to target. Empty selects from the workload's pods using POD_AFFECTED_PERCENTAGE.""
TARGET_CONTAINERContainer in the pod running the JVM. Empty targets the first container in the pod spec.""
NODE_LABELLabel selector to filter target pods by the node they run on. Empty disables node-based filtering.""
POD_AFFECTED_PERCENTAGEPercentage of the workload's pods to target. 0 means one pod.0
SEQUENCEWhen multiple pods are targeted, inject parallel (all at once) or serial (one after another).parallel

Runtime and helper

TunableDescriptionDefault
CONTAINER_RUNTIMEContainer runtime on the target nodes. One of containerd, docker, crio.containerd
SOCKET_PATHPath to the container runtime socket on the target node. Set to match CONTAINER_RUNTIME./run/containerd/containerd.sock
RAMP_TIMEWait period in seconds before and after the fault. Go to ramp time to read how it is applied.0

Common pod selection tunables (TARGET_WORKLOAD_KIND, TARGET_WORKLOAD_NAMESPACE, TARGET_WORKLOAD_NAMES, TARGET_WORKLOAD_LABELS) are documented in common pod fault tunables. Tunables that apply to every fault are documented in common tunables for all faults.

Match mode to scenario

Producer-side latency surfaces as slow send().get() and increased in-flight buffer use. Consumer-side latency surfaces as slow poll() returns and growing lag. Target the mode that matches the failure you want to validate.

Configure for your container runtime​

Set CONTAINER_RUNTIME and SOCKET_PATH to match the runtime on the target node:

CONTAINER_RUNTIMESOCKET_PATH
containerd (default)/run/containerd/containerd.sock
docker/var/run/docker.sock
crio/var/run/crio/crio.sock

Fault execution in brief​

Attaches a Java agent to the target JVM and intercepts Kafka client operations matching KAFKA_MODE and KAFKA_TOPIC to add LATENCY milliseconds to each matched call on the configured percentage, for TOTAL_CHAOS_DURATION seconds.

Expected behavior during fault execution​

  • Matched Kafka operations take longer by approximately LATENCY ms. Other topics and the unmatched mode run normally.
  • Caller-side metrics (request latency, in-flight buffer use, consumer lag) rise to reflect the added delay.
  • Clients with timeouts shorter than LATENCY cancel the call and may retry.
  • Consumer applications may see growing lag if LATENCY per poll exceeds the processing budget.
  • Tracing systems show the matched Kafka span growing by LATENCY ms.
When the fault ends

Kafka calls return to baseline latency immediately. Calls in flight finish at the delayed time and then the system returns to normal.

Signals to watch​

Attach resilience probes to assert each layer:

  • Producer latency: Use a Prometheus probe on kafka_producer_request_latency_avg.
  • Consumer lag: Use a Prometheus probe on kafka_consumer_lag or your dashboard's lag metric.
  • End-to-end latency: Use an HTTP probe against an endpoint that triggers a produce or consume.

Verify the fault execution effect​

While the experiment is running, confirm operations are slower:

  1. Time a request that triggers the matched operation.

    kubectl run -n <namespace> tester --image=nicolaka/netshoot --rm -it -- \
      curl -w "time=%{time_total}\n" -o /dev/null -s http://<service>:<port>/<endpoint>

  2. Confirm in tracing.

    The Kafka span for the matched operation should be approximately LATENCY ms longer than its baseline.

Recovery and cleanup​

  • End of duration: Kafka calls return to baseline latency automatically.
  • Abort the experiment: Stopping the experiment from Chaos Studio triggers the same cleanup path.
  • Lag drainage: Consumer lag built up during the fault may take time to drain. Allow normal processing to catch up, or scale consumers temporarily if needed.

Limitations​

  • Serverless Kubernetes (EKS Fargate, ACI virtual nodes): These platforms do not expose container runtime sockets and reject the privileged access the fault needs. GKE Autopilot is supported once the one-time setup in Chaos on GKE Autopilot is in place.
  • Windows containers: This fault is supported on Linux pods only.
  • Non-JVM and non-Kafka workloads: This fault targets the Apache Kafka Java client inside a JVM.
  • Streams API: Applications using Kafka Streams may attribute latency to internal operators rather than client calls; behavior depends on the topology.

Troubleshooting​

Pod JVM Kafka latency experiment stays Pending or never starts in Harness Chaos Engineering

Inspect the chaos pods in the experiment namespace with kubectl describe pod -n <chaos-namespace>. The most common causes are taints on the target node that the chaos pods do not tolerate, insufficient resources, or a PodSecurity admission policy blocking privileged pods. Add the required tolerations or run in a namespace with privileged Pod Security level.

No latency observed during pod-jvm-kafka-latency

The most common causes are: KAFKA_MODE does not match what the application uses (producer vs consumer); KAFKA_TOPIC does not match the topic the application produces/consumes; TRANSACTION_PERCENTAGE is 0 (default); or the path uses Kafka Streams. Re-run with TRANSACTION_PERCENTAGE=100, empty KAFKA_TOPIC, and a larger LATENCY to confirm.

Connection to container runtime fails for pod-jvm-kafka-latency in Harness Chaos Engineering

The default SOCKET_PATH is /run/containerd/containerd.sock. For Docker, set CONTAINER_RUNTIME=docker and SOCKET_PATH=/var/run/docker.sock. For CRI-O, set CONTAINER_RUNTIME=crio and SOCKET_PATH=/var/run/crio/crio.sock.