bakery-ia/infrastructure/kubernetes/apply-monitoring-to-all.py

#!/usr/bin/env python3
"""
Script to automatically add OpenTelemetry monitoring configuration to all service deployments.
This adds environment variables for metrics, logs, and traces export to SigNoz.
"""

import os
import re
import sys
from pathlib import Path

# Services to configure
SERVICES = [
    "ai-insights",
    "distribution",
    "external",
    "forecasting",
    "inventory",
    "notification",
    "orchestrator",
    "orders",
    "pos",
    "procurement",
    "production",
    "recipes",
    "sales",
    "suppliers",
    "tenant",
    "training",
]

OTEL_ENV_VARS_TEMPLATE = """        env:
        # OpenTelemetry Configuration
        - name: OTEL_COLLECTOR_ENDPOINT
          value: "http://signoz-otel-collector.signoz.svc.cluster.local:4318"
        - name: OTEL_EXPORTER_OTLP_ENDPOINT
          value: "http://signoz-otel-collector.signoz.svc.cluster.local:4318"
        - name: OTEL_SERVICE_NAME
          value: "{service_name}"
        - name: ENABLE_TRACING
          value: "true"
        # Logging Configuration
        - name: OTEL_LOGS_EXPORTER
          value: "otlp"
        - name: OTEL_PYTHON_LOGGING_AUTO_INSTRUMENTATION_ENABLED
          value: "true"
        # Metrics Configuration (all via OTLP, no Prometheus)
        - name: ENABLE_OTEL_METRICS
          value: "true"
        - name: ENABLE_SYSTEM_METRICS
          value: "true"
"""


def has_otel_config(content: str) -> bool:
    """Check if file already has OTEL configuration"""
    return "OTEL_COLLECTOR_ENDPOINT" in content


def add_otel_config(content: str, service_name: str) -> str:
    """Add OTEL configuration to service deployment"""

    # Prepare the env vars with the service name
    env_vars = OTEL_ENV_VARS_TEMPLATE.format(service_name=f"{service_name}-service")

    # Find the container section and add env vars before envFrom
    # Pattern: find "      containers:" then first "        envFrom:" after it
    pattern = r'(      containers:\n      - name: [^\n]+\n        image: [^\n]+\n(?:        ports:\n(?:        - [^\n]+\n)+)?)(        envFrom:)'

    replacement = r'\1' + env_vars + r'\2'

    # Try to replace
    new_content = re.sub(pattern, replacement, content, count=1)

    if new_content == content:
        print(f"  ⚠️  Warning: Could not find insertion point automatically")
        return content

    return new_content


def process_service(service_name: str, base_path: Path) -> bool:
    """Process a single service deployment file"""

    service_file = base_path / "components" / service_name / f"{service_name}-service.yaml"

    if not service_file.exists():
        print(f"  ⚠️  File not found: {service_file}")
        return False

    # Read file
    with open(service_file, 'r') as f:
        content = f.read()

    # Check if already configured
    if has_otel_config(content):
        print(f"  ✓ Already configured")
        return True

    # Add configuration
    new_content = add_otel_config(content, service_name)

    if new_content == content:
        return False

    # Write back
    with open(service_file, 'w') as f:
        f.write(new_content)

    print(f"  ✅ Updated successfully")
    return True


def main():
    """Main function"""

    # Find base path
    script_dir = Path(__file__).parent
    base_path = script_dir / "base"

    if not base_path.exists():
        print(f"❌ Error: Base path not found: {base_path}")
        sys.exit(1)

    print("=" * 60)
    print("Adding OpenTelemetry Monitoring Configuration")
    print("=" * 60)
    print()

    success_count = 0
    skip_count = 0
    fail_count = 0

    for service in SERVICES:
        print(f"📝 Processing {service}-service...")

        result = process_service(service, base_path)

        if result:
            if has_otel_config(open(base_path / "components" / service / f"{service}-service.yaml").read()):
                success_count += 1
        else:
            fail_count += 1

        print()

    print("=" * 60)
    print(f"✅ Successfully configured: {success_count}")
    if fail_count > 0:
        print(f"⚠️  Failed to configure: {fail_count}")
    print("=" * 60)
    print()

    print("Next steps:")
    print("1. Review the changes: git diff infrastructure/kubernetes/base/components/")
    print("2. Update SigNoz: helm upgrade signoz signoz/signoz -n signoz -f infrastructure/helm/signoz-values-dev.yaml")
    print("3. Apply changes: kubectl apply -k infrastructure/kubernetes/overlays/dev/")
    print("4. Verify: kubectl logs -n bakery-ia deployment/<service-name> | grep -i 'otel\\|metrics'")


if __name__ == "__main__":
    main()