Makefile.observability

.PHONY: help observability-stack test test-unit test-chaos test-stress test-soak test-redteam life-plans

help:
	@echo "HyperCode Observability & Resilience Test Suite"
	@echo ""
	@echo "Available commands:"
	@echo "  make observability-stack    Start Prometheus + Loki + Grafana"
	@echo "  make test-unit              Run unit tests (health checks, SLO baselines)"
	@echo "  make test-chaos             Run chaos tests (failure injection + recovery)"
	@echo "  make test-stress            Run stress tests (100 req/sec, memory leaks)"
	@echo "  make test-soak              Run 72-hour soak test (SLOW!)"
	@echo "  make test-redteam           Run red-team tests (security, adversarial)"
	@echo "  make test                   Run all tests (unit + chaos + stress)"
	@echo "  make life-plans             Generate agent life-plan YAMLs"
	@echo "  make dashboards             Provision Grafana dashboards"
	@echo "  make logs                   Tail Docker logs (all services)"
	@echo "  make metrics                Query Prometheus metrics"
	@echo "  make clean                  Stop all containers, clear volumes"
	@echo ""

# ============================================================================
# STACK MANAGEMENT
# ============================================================================

observability-stack:
	@echo "Starting observability stack: Prometheus, Loki, Tempo, Grafana..."
	docker-compose up -d prometheus loki tempo grafana promtail node-exporter cadvisor
	@echo "Waiting for services to be healthy..."
	@sleep 5
	@echo "✓ Prometheus:  http://localhost:9090"
	@echo "✓ Loki:        http://localhost:3100"
	@echo "✓ Grafana:     http://localhost:3001 (admin/admin)"
	@echo "✓ Tempo:       http://localhost:3200"

# ============================================================================
# TESTS
# ============================================================================

test-unit:
	@echo "Running unit tests (health, baselines)..."
	python3 tests/run_tests.py --suite unit

test-chaos:
	@echo "Running chaos tests (failure injection + recovery)..."
	python3 tests/run_tests.py --suite chaos

test-stress:
	@echo "Running stress tests (high load, memory detection)..."
	python3 tests/run_tests.py --suite stress

test-soak:
	@echo "Running 72-hour soak test (TAKES LONG TIME)..."
	python3 tests/run_tests.py --suite soak --soak-hours 72

test-soak-quick:
	@echo "Running quick 1-hour soak test (for testing)..."
	python3 tests/run_tests.py --suite soak --soak-hours 1

test-redteam:
	@echo "Running red-team tests (security, adversarial)..."
	python3 tests/run_tests.py --suite redteam

test:
	@echo "Running full test suite: unit + chaos + stress..."
	python3 tests/run_tests.py --suite all

# ============================================================================
# LIFE PLANS & DASHBOARDS
# ============================================================================

life-plans:
	@echo "Agent life-plans (YAML runbooks) are at:"
	@echo "  agents/life-plans/TEMPLATE.md (guide)"
	@echo "  agents/life-plans/hypercode-core.yaml (example)"
	@echo "  agents/life-plans/crew-orchestrator.yaml (example)"
	@echo ""
	@echo "To create more:"
	@echo "  cp agents/life-plans/TEMPLATE.md agents/life-plans/[agent-name].yaml"
	@echo "  vim agents/life-plans/[agent-name].yaml"
	@echo "  git add agents/life-plans/ && git commit -m 'docs: add life plan for [agent-name]'"

dashboards:
	@echo "Grafana dashboards will be auto-provisioned from:"
	@echo "  monitoring/grafana/provisioning/dashboards/"
	@echo ""
	@echo "Manual access: http://localhost:3001"
	@echo "  Username: admin"
	@echo "  Password: admin"
	@echo ""
	@echo "To add custom dashboard:"
	@echo "  1. Create in Grafana UI"
	@echo "  2. Export as JSON"
	@echo "  3. Save to monitoring/grafana/provisioning/dashboards/[name].json"
	@echo "  4. Restart Grafana: docker-compose up -d grafana"

# ============================================================================
# DEBUGGING & MONITORING
# ============================================================================

logs:
	@echo "Tailing logs from all services..."
	docker-compose logs -f hypercode-core crew-orchestrator redis postgres

logs-service:
	@echo "Usage: make logs-service SERVICE=hypercode-core"
	docker-compose logs -f $(SERVICE)

metrics:
	@echo "Prometheus queries:"
	@echo "  UP status:              curl 'http://localhost:9090/api/v1/query?query=up' | jq"
	@echo "  Request latency:        curl 'http://localhost:9090/api/v1/query?query=http_request_duration_seconds_bucket' | jq"
	@echo "  Error rate:             curl 'http://localhost:9090/api/v1/query?query=http_requests_total' | jq"
	@echo "  Container memory:       curl 'http://localhost:9090/api/v1/query?query=container_memory_usage_bytes' | jq"

alerts:
	@echo "Alert rules:"
	@echo "  Rules file: monitoring/prometheus/alert_rules.yml"
	@echo "  Alertmanager: http://localhost:9093 (if configured)"

status:
	@echo "Service status:"
	docker-compose ps

# ============================================================================
# CLEANUP
# ============================================================================

clean:
	@echo "Stopping all services..."
	docker-compose down
	@echo "Clearing volumes (DESTRUCTIVE - data loss!)..."
	docker volume prune -f

restart:
	@echo "Restarting observability stack..."
	docker-compose restart prometheus loki grafana tempo

# ============================================================================
# CI/CD INTEGRATION
# ============================================================================

ci-test:
	@echo "Running tests in CI environment..."
	python3 tests/run_tests.py --suite all
	@echo "Collecting test results..."
	@if [ -f /tmp/soak_test_metrics.json ]; then echo "✓ Soak metrics saved"; fi

ci-publish-results:
	@echo "Publishing metrics to external system..."
	@echo "TODO: Integrate with Datadog/Splunk/CloudWatch"

# ============================================================================
# LOCAL DEVELOPMENT
# ============================================================================

dev-watch-logs:
	@echo "Watching logs in real-time..."
	docker-compose logs -f --tail=50

dev-health-check:
	@echo "Checking all service health endpoints..."
	@echo "HyperCode Core:     $$(curl -s http://localhost:8000/health | jq '.status' 2>/dev/null || echo 'FAIL')"
	@echo "Crew Orchestrator:  $$(curl -s http://localhost:8081/health | jq '.status' 2>/dev/null || echo 'FAIL')"
	@echo "Prometheus:         $$(curl -s http://localhost:9090/-/healthy | head -1 2>/dev/null || echo 'FAIL')"
	@echo "Grafana:            $$(curl -s http://localhost:3001/api/health | jq '.status' 2>/dev/null || echo 'FAIL')"

dev-scale-agents:
	@echo "Scaling agents (for load testing)..."
	docker-compose up -d --scale backend-specialist=3 --scale frontend-specialist=3

# ============================================================================
# DOCUMENTATION
# ============================================================================

docs:
	@echo "Documentation:"
	@echo "  Life Plans:        agents/life-plans/TEMPLATE.md"
	@echo "  Test Suite:        tests/test_agent_observability.py"
	@echo "  Monitoring Setup:  monitoring/"
	@echo "  This Makefile:     Makefile"

docs-generate:
	@echo "Auto-generating documentation from code..."
	@mkdir -p docs/generated
	pdoc3 --html agents/ -o docs/generated
	@echo "Generated: docs/generated/"