Aller au contenu principal

Troubleshooting

Common issues encountered when running STOA, with diagnosis steps and solutions.

Installation & Deployment​

Pods Not Starting​

Symptom: Pods stuck in CrashLoopBackOff or Pending.

Diagnosis:

kubectl describe pod <pod-name> -n stoa-system
kubectl logs <pod-name> -n stoa-system --previous

Common causes:

CauseFix
Missing secretsCreate required secrets: kubectl create secret generic stoa-db-credentials ...
Insufficient resourcesIncrease node resources or reduce replica count
Image pull failureVerify image exists and pull secret is configured
CRDs not appliedRun kubectl apply -f charts/stoa-platform/crds/

Helm Install Fails​

Symptom: helm upgrade --install returns an error.

# Check Helm release status
helm status stoa-platform -n stoa-system

# View rendered templates
helm template stoa-platform ./charts/stoa-platform -f values.yaml

# Lint the chart
helm lint charts/stoa-platform

Database​

Connection Refused​

Symptom: connection refused or could not connect to server in API logs.

Checklist:

  1. Verify database is running: kubectl get pods -l app=postgresql
  2. Check connection string: DATABASE_URL must include correct host, port, credentials
  3. Check network policies: ensure stoa-system namespace can reach the database
  4. Check security groups (AWS): RDS must allow inbound from EKS nodes

Migration Failures​

Symptom: API starts but returns 500 errors on data operations.

cd control-plane-api
alembic upgrade head
alembic current  # Verify migration state

If migrations are stuck:

alembic stamp head  # Reset to current state (use cautiously)

Authentication​

Configure your environment

The examples below use environment variables. Set them for your STOA instance:

export STOA_API_URL="https://api.gostoa.dev"       # Replace with your domain
export STOA_AUTH_URL="https://auth.gostoa.dev"      # Keycloak OIDC provider
export STOA_GATEWAY_URL="https://mcp.gostoa.dev"    # MCP Gateway endpoint
export STOA_CONSOLE_URL="https://console.gostoa.dev"

Self-hosted? Replace gostoa.dev with your domain.

401 Unauthorized on All Requests​

Checklist:

  1. Token expired? Re-authenticate: stoa login --server ...
  2. Keycloak reachable? curl ${STOA_AUTH_URL}/health
  3. Realm correct? Must be stoa
  4. Client ID correct? control-plane-ui for Console, stoa-portal for Portal
  5. Clock skew? Ensure server and client clocks are synchronized

CORS Errors During Login​

Symptom: Browser console shows CORS errors when redirecting to/from Keycloak.

Fix: In Keycloak admin:

  1. Open the client configuration (control-plane-ui or stoa-portal)
  2. Add the application URL to Web Origins (e.g., ${STOA_CONSOLE_URL})
  3. Add to Valid Redirect URIs (e.g., ${STOA_CONSOLE_URL}/*)

Token Missing Scopes​

Symptom: 403 Forbidden despite being authenticated.

Fix: Ensure the user belongs to the correct Keycloak group:

RoleGroupScopes
Platform admincpi-adminstoa:admin, stoa:write, stoa:read
Tenant admintenant-adminstoa:write, stoa:read
DevOpsdevopsstoa:write, stoa:read
Viewerviewerstoa:read

MCP Gateway​

Tools Not Appearing​

Symptom: tools/list returns empty array.

Checklist:

  1. CRDs applied? kubectl get crds | grep gostoa.dev
  2. Tools created? kubectl get tools -A
  3. Watcher enabled? Set K8S_WATCHER_ENABLED=true
  4. Correct namespace? Tools must be in tenant-{name} namespaces
  5. Gateway restarted after CRD changes? (if watcher is disabled)

OPA Policy Denying All Requests​

Symptom: All tools/call requests return 403.

# Check OPA is enabled and configured
kubectl logs -l app=mcp-gateway -n stoa-system | grep opa

# Verify policy files are mounted
kubectl exec -it <mcp-gateway-pod> -n stoa-system -- ls /policies/

Verify your Rego policy has a path to allow = true for your use case.

Kafka Metering Errors​

Symptom: MCP Gateway logs show Kafka connection errors.

Checklist:

  1. Kafka/Redpanda running? kubectl get pods -l app=redpanda
  2. Bootstrap servers correct? Check KAFKA_BOOTSTRAP_SERVERS
  3. Topic exists? kubectl exec redpanda-0 -- rpk topic list
  4. Network policies allow traffic? Check cross-namespace rules

Portal​

APIs Not Showing in Catalog​

Symptom: Portal loads but catalog is empty.

Checklist:

  1. APIs published? Check in Console UI or via API
  2. Portal visibility enabled? API must have portal.visible: true
  3. API endpoint returns data? curl ${STOA_API_URL}/v1/portal/apis
  4. Auth working? Portal needs valid token to fetch catalog

Search Returns 500​

Symptom: Searching in Portal returns HTTP 500.

This was a known issue (CAB-1044) caused by unescaped LIKE wildcards. Ensure you're running a version with the fix (commit 2c5672d8 or later).

Gateway Sync​

API Not Synced to Gateway​

Symptom: API shows as "Published" in Console but not reachable on the gateway.

Checklist:

  1. ArgoCD sync working? Check ArgoCD application status
  2. Gateway adapter healthy? curl ${STOA_API_URL}/v1/gateways — check status: online
  3. Deployment sync status? Check sync_status in Console deployment view
  4. Gateway credentials valid? Verify gateway instance auth_config in the Control Plane

Drift Detected​

Symptom: Console shows "Drift" status for an API.

Drift means the gateway state differs from the desired state in the Control Plane. The reconciliation engine detects this automatically. To trigger a manual re-sync:

# Via Control Plane API
curl -X POST ${STOA_API_URL}/v1/deployments/{deployment_id}/sync \
  -H "Authorization: Bearer $TOKEN"

# Check deployment sync status
curl ${STOA_API_URL}/v1/deployments/{deployment_id} \
  -H "Authorization: Bearer $TOKEN"

Vault​

Vault Sealed​

Symptom: Application returns errors related to Vault, logs show VaultSealedException.

Fix: Unseal Vault using the unseal keys:

vault operator unseal <key1>
vault operator unseal <key2>
vault operator unseal <key3>

This was addressed in CAB-1042 with automatic sealed detection (_ensure_unsealed()).

Performance​

High API Latency​

  1. Check database query times: Grafana → Control Plane API dashboard
  2. Check Kafka consumer lag: kubectl exec redpanda-0 -- rpk group describe stoa-events
  3. Check pod resource usage: kubectl top pods -n stoa-system
  4. Review slow query logs: Loki → {app="control-plane-api"} | json | duration > 1s

High Memory Usage​

  1. Check for connection pool leaks: Monitor DB_POOL_SIZE usage in Prometheus
  2. Review Kafka consumer configuration: Reduce batch sizes if needed
  3. Check for large API specifications being loaded into memory