cloudbuild.yaml

steps:

  # --- STEP 1: Build and Push Docker Image ---
  - name: 'gcr.io/cloud-builders/docker'
    args:
      - 'build'
      - '-t'
      - 'us-central1-docker.pkg.dev/$PROJECT_ID/toolbox-evals/cloud-sql-postgresql:latest'
      - '.'

  - name: 'gcr.io/cloud-builders/docker'
    args:
      - 'push'
      - 'us-central1-docker.pkg.dev/$PROJECT_ID/toolbox-evals/cloud-sql-postgresql:latest'

  # --- STEP 2: Deploy to Cloud Run ---
  - name: 'gcr.io/google.com/cloudsdktool/cloud-sdk'
    entrypoint: gcloud
    args:
      - 'run'
      - 'deploy'
      - 'cloud-sql-postgresql-server'
      - '--image=us-central1-docker.pkg.dev/$PROJECT_ID/toolbox-evals/cloud-sql-postgresql:latest'
      - '--region=us-central1'
      - '--allow-unauthenticated'
      - '--port=8080'
      - '--timeout=300'
      - '--set-env-vars=CLOUD_SQL_POSTGRES_PROJECT=$PROJECT_ID,CLOUD_SQL_POSTGRES_INSTANCE=omkar-demo-postgres-1,CLOUD_SQL_POSTGRES_REGION=us-central1,CLOUD_SQL_POSTGRES_DATABASE=postgres,CLOUD_SQL_POSTGRES_USER=postgres,CLOUD_SQL_POSTGRES_PASSWORD=[PASSWORD],CLOUD_SQL_POSTGRES_IP_TYPE=PUBLIC'

  # --- STEP 3: Fully Integrated Evaluation to Persist Results ---
  - name: 'us-central1-docker.pkg.dev/$PROJECT_ID/toolbox-evals/eval_server:latest'
    entrypoint: 'bash'
    args:
      - '-c'
      - |
        set -e
        cd /evalbench

        export EVAL_GCP_PROJECT_ID=$PROJECT_ID
        export EVAL_GCP_PROJECT_REGION=us-central1

        echo "Compiling protobuf files..."
        python3 -m grpc_tools.protoc --proto_path=evalbench/evalproto --python_out=evalbench/evalproto --grpc_python_out=evalbench/evalproto evalbench/evalproto/*.proto

        echo "Patching client to use insecure credentials..."
        # sed -i 's/"localhost:50051"/"127.0.0.1:50051"/g' evalbench/client/eval_client.py
        sed -i 's/grpc.alts_channel_credentials()/None/g' evalbench/client/eval_client.py
        sed -i 's/grpc.aio.secure_channel(address, channel_creds)/grpc.aio.insecure_channel(address)/g' evalbench/client/eval_client.py

        echo "Patching server to listen on all IPv4 interfaces (0.0.0.0)..."
        sed -i 's/"\[::\]:%s"/"0.0.0.0:%s"/g' /evalbench/evalbench/eval_server.py
        echo "Checking bind success in server (writing to stderr)..."
        sed -i 's|server.add_insecure_port("0.0.0.0:%s" % PORT)|bound_port = server.add_insecure_port("0.0.0.0:%s" % PORT)\n        import sys\n        sys.stderr.write(f"BOUND_PORT: {bound_port}\\n")\n        if bound_port == 0: raise RuntimeError("Failed to bind to port!")|' /evalbench/evalbench/eval_server.py

        echo "Patching eval_service.py to fix TypeError in get_reporters..."
        sed -i 's|reporters = get_reporters(config.get("reporting"), job_id, run_time)|reporters = get_reporters(config.get("reporting") or {}, job_id, run_time)|' /evalbench/evalbench/eval_service.py

        echo "Patching util/session.py to make ADK import lazy..."
        sed -i 's|from google.adk.sessions import VertexAiSessionService||' /evalbench/evalbench/util/session.py
        sed -i 's|    def __init__(self, config):|    def __init__(self, config):\n        from google.adk.sessions import VertexAiSessionService|' /evalbench/evalbench/util/session.py
        echo "Patching databases/util.py to make SecretManagerClient lazy..."
        sed -i 's|CLIENT = secretmanager_v1.SecretManagerServiceClient()|CLIENT = None\ndef get_client():\n    global CLIENT\n    if CLIENT is None:\n        CLIENT = secretmanager_v1.SecretManagerServiceClient()\n    return CLIENT|' /evalbench/evalbench/databases/util.py || echo "Failed to patch databases/util.py"
        sed -i 's|CLIENT.access_secret_version|get_client().access_secret_version|' /evalbench/evalbench/databases/util.py || echo "Failed to patch databases/util.py usage"
        cd evalbench
        export PROTOCOL_BUFFERS_PYTHON_IMPLEMENTATION=python
        export PYTHONPATH=./evalproto:.
        export CLOUD_RUN=True
        export PORT=50051



        echo "Starting Evaluation Server in background..."
        # NEW: Added </dev/null in case it was waiting for input
        python3 -u ./eval_server.py --localhost </dev/null &
        SERVER_PID=$$!
        
        echo "Waiting for port 50051 to open..."
        python3 -c "
        import socket
        import time
        for i in range(20):
            try:
                s = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
                s.connect(('127.0.0.1', 50051))
                print('Port is open!')
                exit(0)
            except Exception as e:
                print(f'Port not open yet: {e}')
                time.sleep(1)
        print('Port failed to open')
        exit(1)
        " || { echo "Server failed to bind port. Check logs above."; exit 1; }
        
        echo "Server is running. Launching Evaluation Client..."
        cd /evalbench
        export PYTHONPATH=./evalbench:./evalbench/evalproto
        
        python3 evalbench/client/eval_client.py --experiment=/workspace/evals/run_config.yaml --endpoint=local || { echo "Client failed! Server logs:"; cat /evalbench/evalbench/server.log; exit 1; }