Skip to main content
This page lists common failure modes for the chart with their diagnostic and recovery steps.

​
Engine never becomes Ready: “local file system for managed storage”

The engine logs: 
Invalid Firebolt configuration: local file system for managed storage
is not supported when using dedicated Pensieve
customEngineConfig.storage is unset. Without object storage the engine does not enter the Ready state. Configure object storage and run helm upgrade. See Object Storage. Confirm the failure is the managed-storage check: 
# Search the engine pod's logs for the managed-storage error.
kubectl -n firebolt logs <engine-pod> | grep -i "managed storage"

​
Engine CrashLoopBackOff: metadata RPC failure at startup

Engine logs show a gRPC error connecting to the Metadata Service. The most common cause is an engine-metadata image-version mismatch. The two images are not version-tolerant of each other. Inspect the rendered tags: 
# Print the engine pods' image tags.
kubectl -n firebolt get pods -l firebolt/component=engine \
  -o jsonpath='{range .items[*]}{.metadata.name}{"\t"}{.spec.containers[*].image}{"\n"}{end}'

# Same for the Metadata Service pods.
kubectl -n firebolt get pods -l firebolt/component=metadata-service \
  -o jsonpath='{range .items[*]}{.metadata.name}{"\t"}{.spec.containers[*].image}{"\n"}{end}'

# Print the release's rendered image values to compare engine and metadata tags.
helm -n firebolt get values firebolt | grep -E 'tag:|repository:'
Keep engineSpec.image.tag and metadata.image.tag in lockstep. By default both fall back to Chart.yaml’s appVersion, which is correct. See Image overrides.

​
Gateway returns 404 or “no upstream”

The X-Firebolt-Engine header value does not match any name under engines:. The Envoy Lua filter also rejects names that are not RFC 1123 DNS labels (lowercase alphanumerics and hyphens, max 63 chars) with a 400. Confirm what engines the release knows about: 
# Print the engines: list from the release's rendered values.
helm -n firebolt get values firebolt | grep -A1 'engines:'
Use a matching engine name, or add the engine to engines: and helm upgrade.

​
Engine pod stuck Pending

The PVC is unbound. The chart provisions PVCs without storageClassName and falls back to the cluster default StorageClass. Inspect the PVCs and the cluster’s storage classes: 
# Show the engine PVCs and the events explaining why they are not bound.
kubectl -n firebolt describe pvc -l firebolt/component=engine

# List storage classes. Look for one marked (default).
kubectl get storageclass
Either mark a StorageClass as default (storageclass.kubernetes.io/is-default-class: "true"), or set an explicit class: 
engineSpec:
  defaultStorage:
    storageClassName: gp3

​
ImagePullBackOff from a private registry

The kubelet cannot authenticate to the registry that hosts the engine or metadata image. The chart does not create pull secrets. Reference one explicitly: 
imagePullSecrets:
  - name: my-registry-creds
Create the Secret in the release namespace: 
# Create a docker-registry-type Secret matching the imagePullSecrets name.
kubectl create secret docker-registry my-registry-creds -n firebolt \
  --docker-server=registry.example.com \
  --docker-username=… --docker-password=…
See Image overrides.

​
Pre-commit hook aborts the commit on first run

When editing the chart, the helm-docs pre-commit hook regenerates helm/README.md and stages the result. If the regeneration changes the file, the commit aborts so you can re-run git commit with the now-staged README included. This is intentional. Re-running the same commit succeeds.