wehub-resource-sync
4b6817381b
5.1 KiB
Chaos engineering + Datadog (kind)
Defaults: cluster tracer-k8s-test · context kind-tracer-k8s-test · Datadog Helm namespace tracer-test.
This directory holds Kubernetes manifests (chaos-demo.yaml, pod-kill-demo.yaml, experiments/<name>/), JSON samples for opensre investigate -i, and the Python helpers behind make chaos-lab-* / make chaos-experiment-* (python -m tests.chaos_engineering). Alert paths in JSON use the repo-relative form tests/chaos_engineering/experiments/.../foo-alert.json.
Quick start (recommended)
From repo root: Docker, kind, kubectl, helm, and the project deps installed via uv (make install).
export DD_API_KEY='your-api-key' # omit if using CHAOS_LAB_FLAGS=--skip-datadog
# export KUBECTL_CONTEXT=kind-tracer-k8s-test # optional
make chaos-lab-up
make chaos-experiment-list
make chaos-experiment-up EXPERIMENT=pod-failure
make chaos-experiment-down EXPERIMENT=pod-failure
make chaos-lab-down
Flags (via Make): CHAOS_LAB_FLAGS='--skip-datadog', '--skip-kind', '--no-wait-datadog'. Teardown: CHAOS_LAB_DOWN_FLAGS='--keep-kind' or '--keep-datadog'.
Debugging (same as Make): python -m tests.chaos_engineering lab up, experiment list, etc.
Convention: Each experiment uses sorted *-demo.yaml then *-chaos.yaml. On delete, chaos CRs first (make chaos-experiment-down).
Baselines: make chaos-engineering-apply applies chaos-demo, crashloop workload, and pod-kill-demo PodChaos only. Other scenarios use make chaos-experiment-up EXPERIMENT=<name> or raw kubectl (workload before chaos; delete chaos before workload if doing it manually).
Discover experiments: make chaos-experiment-list (or browse experiments/). Examples include network faults (network-delay, network-partition, …), crashloop, pod-failure, stress-cpu, dns-error, http-abort, and others.
| Example directory | Chaos Mesh (illustrative) |
|---|---|
experiments/network-delay/ |
NetworkChaos delay |
experiments/crashloop/ |
CrashLoopBackOff + optional PodChaos |
experiments/pod-failure/ |
PodChaos pod-failure (pause image) |
experiments/dns-error/ |
DNSChaos (needs Chaos DNS Server) |
Manual setup (without make chaos-lab-up)
Use this if you want to run the same steps by hand. Sections 1–2 mirror what chaos-lab-up does before Chaos Mesh and baselines.
1. Cluster
kind create cluster --name tracer-k8s-test --wait 60s
kubectl config use-context kind-tracer-k8s-test
2. Datadog
export DD_API_KEY='your-api-key'
# export DD_SITE=datadoghq.eu # if not US1
kubectl create namespace tracer-test --context=kind-tracer-k8s-test --dry-run=client -o yaml \
| kubectl apply -f - --context=kind-tracer-k8s-test
helm repo add datadog https://helm.datadoghq.com
helm repo update datadog
helm upgrade --install datadog datadog/datadog \
--kube-context kind-tracer-k8s-test \
-n tracer-test \
-f tests/e2e/kubernetes/k8s_manifests/datadog-values.yaml \
--set "datadog.apiKey=${DD_API_KEY}" \
${DD_SITE:+--set datadog.site=${DD_SITE}} \
--wait --timeout 10m
Confirm a node Agent exists (kubectl get daemonset,pods -n tracer-test --context=kind-tracer-k8s-test). If there is no agent DaemonSet, fix datadog-values.yaml (e.g. datadog.operator.enabled: false) and upgrade again.
3. Chaos Mesh + baseline workloads
make chaos-mesh-up
kubectl get pods -n chaos-mesh --context=kind-tracer-k8s-test
make chaos-engineering-apply
kubectl get pods -n default --context=kind-tracer-k8s-test
Optional extra PodChaos on crashloop: make chaos-experiment-up EXPERIMENT=crashloop, or apply experiments/crashloop/pod-kill-crashloop-chaos.yaml after the crashloop demo workload is up.
Pod-failure (not part of chaos-engineering-apply): make chaos-experiment-up EXPERIMENT=pod-failure, or apply experiments/pod-failure/*-demo.yaml then *-chaos.yaml. Expect 0/1 READY while chaos is active (pause image). Teardown: delete chaos CR first.
4. Datadog logs
After the crashloop pod runs, wait a few minutes, then query logs e.g. kube_cluster_name:tracer-k8s-test kube_namespace:default crashloop-demo or kube_deployment:crashloop-demo.
5. OpenSRE
opensre onboard # Datadog keys as needed
opensre investigate -i tests/chaos_engineering/experiments/crashloop/crashloop-demo-alert.json
Use the matching *-alert.json under experiments/<name>/ for other scenarios. Local kind is observed via Datadog (or a richer synthetic payload), not EKS APIs.
Cleanup
make chaos-engineering-delete
make chaos-mesh-down
helm uninstall datadog -n tracer-test --kube-context kind-tracer-k8s-test
kubectl delete namespace tracer-test --context=kind-tracer-k8s-test
kind delete cluster --name tracer-k8s-test
Or use make chaos-lab-down (and flags) if you brought the lab up with chaos-lab-up.
Make targets: chaos-lab-up, chaos-lab-down, chaos-experiment-list, chaos-experiment-up, chaos-experiment-down, chaos-mesh-up, chaos-mesh-down, chaos-engineering-apply, chaos-engineering-delete. Override context with KUBECTL_CONTEXT=... when needed.