Production Debugging

Systematic approach to debugging production issues in Kubernetes microservice environments.

When to Use

• Investigating HTTP 500 errors
• Debugging missing functionality (feature works locally, fails in production)
• Tracing requests across microservices
• Finding silent failures (no error, but wrong behavior)
• Service-to-service integration issues

Debugging Methodology

Step 1: Reproduce and Identify Symptoms

# What's the user seeing?
# - HTTP 500 error on /workers page
# - No reminder notifications
# - Data saved but logs show errors

# Document the symptom precisely before diving in

Step 2: Check Logs Systematically

# Start with the failing service
kubectl logs deploy/<service-name> -n <namespace> --tail=100

# Filter for errors
kubectl logs deploy/<service-name> -n <namespace> --tail=200 | grep -i -E "(error|exception|fail|warn)"

# Check specific container in multi-container pod
kubectl logs deploy/<service-name> -n <namespace> -c <container-name> --tail=100

# Common containers:
# - main app container (e.g., "api", "web")
# - daprd (Dapr sidecar)
# - init containers (e.g., "wait-for-db")

Step 3: Trace the Request Path

For microservice issues, trace the full request path:

# 1. Frontend → API
kubectl logs deploy/web-dashboard -n taskflow --tail=50

# 2. API processing
kubectl logs deploy/taskflow-api -n taskflow --tail=100 | grep -i "endpoint-name"

# 3. API → External service (e.g., Dapr, SSO)
kubectl logs deploy/taskflow-api -n taskflow -c daprd --tail=50

# 4. Downstream service
kubectl logs deploy/notification-service -n taskflow --tail=50

Step 4: Analyze the Error

Common patterns to look for:

| Error Pattern | Likely Cause | |---------------|--------------| | AttributeError: 'X' has no attribute 'Y' | Model/schema mismatch | | 404 Not Found on internal call | Wrong endpoint URL | | greenlet_spawn has not been called | Async SQLAlchemy pattern issue | | event_type: None | Message format/unwrapping issue | | Times off by hours | Timezone handling bug |

Quick Commands

Check All Services Status

kubectl get pods -n taskflow
kubectl get pods -n taskflow -o wide  # With node info

Check Service Logs

# Main app logs
kubectl logs deploy/taskflow-api -n taskflow --tail=100

# Dapr sidecar logs
kubectl logs deploy/taskflow-api -n taskflow -c daprd --tail=100

# Follow logs in real-time
kubectl logs deploy/taskflow-api -n taskflow -f

# Logs from specific time
kubectl logs deploy/taskflow-api -n taskflow --since=5m

Check Pod Events

kubectl describe pod <pod-name> -n taskflow
kubectl get events -n taskflow --sort-by='.lastTimestamp'

Execute Commands in Pod

# Shell into pod
kubectl exec -it deploy/taskflow-api -n taskflow -- /bin/sh

# Run specific command
kubectl exec deploy/taskflow-api -n taskflow -- env | grep DATABASE

Common Bug Patterns

1. Model/Schema Mismatch

Symptom: AttributeError: 'Model' has no attribute 'field'

Debug:

# Find the error
kubectl logs deploy/taskflow-api -n taskflow --tail=100 | grep -i "attribute"

# Check the model definition
grep -r "class Worker" apps/api/src/

Fix: Ensure code references match actual model fields.

2. Wrong Endpoint URL

Symptom: 404 Not Found on internal service calls

Debug:

# Check what URL is being called
kubectl logs deploy/taskflow-api -n taskflow -c daprd --tail=100 | grep "404"

# Check what endpoints exist
kubectl exec deploy/taskflow-api -n taskflow -- curl localhost:8000/openapi.json | jq '.paths | keys'

Fix: Match the callback URL to what the service exposes.

3. Timezone Bugs

Symptom: Scheduled jobs fire at wrong times (hours off)

Debug:

# Check when job was scheduled vs when it should fire
kubectl logs deploy/taskflow-api -n taskflow | grep -i "scheduled"

# Compare times
# If local time 23:00 but scheduled for 23:00 UTC → timezone bug

Fix: Convert to UTC before storing/scheduling.

4. Message Format Issues

Symptom: Handler receives data but can't find expected fields

Debug:

# Add logging to see raw message
kubectl logs deploy/notification-service -n taskflow | grep -i "raw"

# Check message structure
# CloudEvent wraps payload in "data" field

Fix: Unwrap CloudEvent: event = raw.get("data", raw)

5. Async SQLAlchemy Errors

Symptom: greenlet_spawn has not been called

Debug:

# Find the line that crashes
kubectl logs deploy/notification-service -n taskflow | grep -A 20 "greenlet"

Fix: Add await session.refresh(obj) after commit before accessing attributes.

Debugging Dapr Specifically

Check Dapr Sidecar

# Dapr scheduler connection
kubectl logs deploy/taskflow-api -n taskflow -c daprd | grep -i "scheduler"

# Dapr API calls
kubectl logs deploy/taskflow-api -n taskflow -c daprd | grep "HTTP API Called"

# Dapr pub/sub
kubectl logs deploy/taskflow-api -n taskflow -c daprd | grep -i "publish"

Check Dapr Subscriptions

# What subscriptions are registered?
kubectl exec deploy/notification-service -n taskflow -- curl localhost:8001/dapr/subscribe

Test Dapr Pub/Sub

# Publish test event from inside cluster
kubectl exec deploy/taskflow-api -n taskflow -- curl -X POST \
  http://localhost:3500/v1.0/publish/taskflow-pubsub/test-topic \
  -H "Content-Type: application/json" \
  -d '{"test": true}'

Debugging Checklist

When investigating a production issue:

• [ ] Reproduce the issue (what exactly fails?)
• [ ] Check pod status (kubectl get pods)
• [ ] Check main app logs for errors
• [ ] Check sidecar logs (daprd, etc.)
• [ ] Trace request path across services
• [ ] Identify error pattern (see table above)
• [ ] Verify fix locally before deploying
• [ ] Deploy and verify in production

CI/CD Integration

Check Deployment Status

# GitHub Actions
gh run list --limit 5

# Check specific run
gh run view <run-id>

# Watch deployment
gh run watch

Verify Deployment

# Check pod restart count (should be 0 for healthy pods)
kubectl get pods -n taskflow

# Check pod age (recent = just deployed)
kubectl get pods -n taskflow -o wide

# Verify new code is running
kubectl logs deploy/taskflow-api -n taskflow --tail=10 | head -5

Prevention

Add Logging at Key Points

logger.info("[SERVICE] Received request: %s", request_summary)
logger.info("[SERVICE] Processing: step=%s, data=%s", step, safe_data)
logger.info("[SERVICE] Completed: result=%s", result_summary)
logger.error("[SERVICE] Failed: error=%s, context=%s", error, context)

Include Correlation IDs

import uuid

@router.post("/tasks")
async def create_task(request: Request):
    correlation_id = request.headers.get("X-Correlation-ID", str(uuid.uuid4()))
    logger.info("[%s] Creating task", correlation_id)
    # ... processing ...
    logger.info("[%s] Task created: %d", correlation_id, task.id)

Test Error Paths

def test_handles_missing_field():
    """Ensure graceful handling of missing data."""
    response = client.post("/tasks", json={})  # Missing required field
    assert response.status_code == 422  # Not 500!