feat&fix: update environment configuration for Docker compatibility and enhance result path handling

monitor/.env

@@ -2,10 +2,13 @@
 # Do not write any secret keys or sensitive information here.
 
 # Monitor configuration
-TASK_CONFIG_PATH=../evaluation_examples/test_small.json
+TASK_CONFIG_PATH=../evaluation_examples/test_all.json
 EXAMPLES_BASE_PATH=../evaluation_examples/examples
-RESULTS_BASE_PATH=../results_operator_aws/pyautogui/screenshot/computer-use-preview
-MAX_STEPS=50
+RESULTS_BASE_PATH=../results_operator_aws
+ACTION_SPACE=pyautogui
+OBSERVATION_TYPE=screenshot
+MODEL_NAME=computer-use-preview
+MAX_STEPS=150
 FLASK_PORT=80
 FLASK_HOST=0.0.0.0
-FLASK_DEBUG=false
+FLASK_DEBUG=true

monitor/docker-compose.yml

@@ -1,5 +1,3 @@
-version: '3'
-
 services:
   monitor:
     build:
@@ -9,10 +7,11 @@ services:
       - "${FLASK_PORT:-8080}:8080"
     volumes:
       - .:/app/monitor
-      - ../evaluation_examples:/app/evaluation_examples
-      - ../results_operator_aws:/app/results_operator_aws
+      - ${TASK_CONFIG_PATH:-../evaluation_examples/test_all.json}:/app/evaluation_examples/test.json
+      - ${EXAMPLES_BASE_PATH:-../evaluation_examples/examples}:/app/evaluation_examples/examples
+      - ${RESULTS_BASE_PATH:-../results_operator_aws}:/app/results
     env_file:
       - .env
     environment:
-      - FLASK_ENV=production
+      - MONITOR_IN_DOCKER=true
     restart: unless-stopped

monitor/main.py

@@ -17,12 +17,26 @@ TASK_STATUS_CACHE = {}
 
 app = Flask(__name__)
 
-# Load configuration from environment variables
-TASK_CONFIG_PATH = os.getenv("TASK_CONFIG_PATH", "../evaluation_examples/test_small.json")
-EXAMPLES_BASE_PATH = os.getenv("EXAMPLES_BASE_PATH", "../evaluation_examples/examples")
-RESULTS_BASE_PATH = os.getenv("RESULTS_BASE_PATH", "../results_operator_aws/pyautogui/screenshot/computer-use-preview")
+MONITOR_IN_DOCKER = os.getenv("MONITOR_IN_DOCKER", "false").lower() == "true"
+
+if MONITOR_IN_DOCKER:
+    # If running in Docker, use default paths
+    TASK_CONFIG_PATH = "/app/evaluation_examples/test.json"
+    EXAMPLES_BASE_PATH = "/app/evaluation_examples/examples"
+    RESULTS_BASE_PATH = "/app/results"
+else:
+    # Load configuration from environment variables
+    TASK_CONFIG_PATH = os.getenv("TASK_CONFIG_PATH", "../evaluation_examples/test_small.json")
+    EXAMPLES_BASE_PATH = os.getenv("EXAMPLES_BASE_PATH", "../evaluation_examples/examples")
+    RESULTS_BASE_PATH = os.getenv("RESULTS_BASE_PATH", "../results")
+
+ACTION_SPACE=os.getenv("ACTION_SPACE", "pyautogui")
+OBSERVATION_TYPE=os.getenv("OBSERVATION_TYPE", "screenshot")
+MODEL_NAME=os.getenv("MODEL_NAME", "computer-use-preview")
 MAX_STEPS = int(os.getenv("MAX_STEPS", "50"))
 
+RESULTS_PATH = os.path.join(RESULTS_BASE_PATH, ACTION_SPACE, OBSERVATION_TYPE, MODEL_NAME)
+
 def load_task_list():
     with open(TASK_CONFIG_PATH, 'r') as f:
         return json.load(f)
@@ -35,7 +49,7 @@ def get_task_info(task_type, task_id):
     return None
 
 def get_task_status(task_type, task_id):
-    result_dir = os.path.join(RESULTS_BASE_PATH, task_type, task_id)
+    result_dir = os.path.join(RESULTS_PATH, task_type, task_id)
     
     if not os.path.exists(result_dir):
         return {
@@ -167,7 +181,7 @@ def get_task_status_brief(task_type, task_id):
     if cache_key in TASK_STATUS_CACHE:
         return TASK_STATUS_CACHE[cache_key]
     
-    result_dir = os.path.join(RESULTS_BASE_PATH, task_type, task_id)
+    result_dir = os.path.join(RESULTS_PATH, task_type, task_id)
     
     if not os.path.exists(result_dir):
         return {
@@ -367,7 +381,7 @@ def api_tasks_brief():
 @app.route('/task/<task_type>/<task_id>/screenshot/<path:filename>')
 def task_screenshot(task_type, task_id, filename):
     """Get task screenshot"""
-    screenshot_path = os.path.join(RESULTS_BASE_PATH, task_type, task_id, filename)
+    screenshot_path = os.path.join(RESULTS_PATH, task_type, task_id, filename)
     if os.path.exists(screenshot_path):
         return send_file(screenshot_path, mimetype='image/png')
     else:
@@ -376,7 +390,7 @@ def task_screenshot(task_type, task_id, filename):
 @app.route('/task/<task_type>/<task_id>/recording')
 def task_recording(task_type, task_id):
     """Get task recording video"""
-    recording_path = os.path.join(RESULTS_BASE_PATH, task_type, task_id, "recording.mp4")
+    recording_path = os.path.join(RESULTS_PATH, task_type, task_id, "recording.mp4")
     if os.path.exists(recording_path):
         response = send_file(recording_path, mimetype='video/mp4')
         # Add headers to improve mobile compatibility