RCA Kind

Root cause analysis workflow for failures on local Kind clusters.

When to Use

Auto-approved: All read and debug operations on Kind clusters are auto-approved.

code

Failure → Reproduce locally → Inspect cluster → Identify root cause → Fix → Verify

Create working directory for analysis:

bash

mkdir -p /tmp/kagenti/rca

Deploy the cluster if not running:

bash

./.github/scripts/local-setup/kind-full-test.sh --skip-cluster-destroy

Check pod status:

bash

kubectl get pods -n kagenti-system

Check for crashes:

bash

kubectl get pods -n kagenti-system --field-selector=status.phase!=Running

Check events:

bash

kubectl get events -n kagenti-system --sort-by='.lastTimestamp'

Check component logs:

bash

kubectl logs -n kagenti-system deployment/kagenti-ui --tail=50

Check Ollama (Kind-specific):

bash

kubectl logs -n kagenti-system deployment/ollama --tail=50

Check agent namespace:

bash

kubectl get pods -n team1

bash

kubectl logs -n team1 deployment/weather-service --tail=50

After fixing, re-run the specific failing test:

bash

uv run pytest kagenti/tests/e2e/ -v -k "test_name"

Issue	Cause	Fix
Ollama OOM	Model too large for Kind	Use smaller model or increase Docker memory
DNS resolution	CoreDNS not ready	Wait or restart CoreDNS pod
Port conflicts	8080 already in use	`lsof -i :8080` and kill process
Image pull errors	Local registry not configured	Check `kind-registry` container is running

If the issue can't be reproduced locally, escalate: