Tutorial: Production Hardening

This guide covers baseline settings for container or VM deployment.

Recommended baseline

CLAUDE_API_KEY=change-me \
CLAUDE_API_TIMEOUT=120 \
CLAUDE_API_MAX_CONCURRENT=4 \
CLAUDE_API_MAX_QUEUE=100 \
CLAUDE_API_QUEUE_TIMEOUT_MS=15000 \
CLAUDE_API_DRAIN_TIMEOUT_SECONDS=30 \
CLAUDE_API_RATE_LIMIT_WINDOW_SECONDS=60 \
CLAUDE_API_RATE_LIMIT_MAX_REQUESTS=30 \
CLAUDE_API_STRICT_HEALTH=true \
npx clauditorium

Graceful shutdown behavior

On SIGTERM/SIGINT:

1. Service enters drain mode.
2. New POST /ask and POST /chat requests return 503 shutting_down.
3. In-flight/queued Claude work is allowed to finish (up to drain timeout).
4. Process exits.

Health and readiness checks

• Liveness/readiness endpoint: GET /health
• Historical readiness data: GET /health/history
• Manual readiness refresh: POST /health/recheck (requires API key)

Observability

• Structured JSON logs for request outcomes and shutdown events
• Prometheus metrics: GET /metrics

Conversation memory safety

Use TTL and max conversation settings to bound memory:

• CLAUDE_API_CONVERSATION_TTL_SECONDS
• CLAUDE_API_MAX_CONVERSATIONS