WIP

app.py

@@ -199,29 +199,41 @@ def download_youtube_audio(url):
 
 def check_api_health():
     """Check if the API is healthy before making requests."""
-    try:
-        response = requests.get(f"{API_URL}/health")
-        response.raise_for_status()
-        health_data = response.json()
-        
-        # Check if service is healthy
-        if health_data.get("status") != "healthy":
-            raise gr.Error("API service is not healthy. Please try again later.")
+    max_retries = 5
+    retry_delay = 10  # seconds
+    last_error = None
+    
+    for attempt in range(max_retries):
+        try:
+            logger.info(f"Performing API health check (attempt {attempt + 1}/{max_retries})...")
+            response = requests.get(f"{API_URL}/health")
+            response.raise_for_status()
+            health_data = response.json()
             
-        # Check resource usage
-        cpu_percent = health_data.get("cpu_percent", 0)
-        memory_percent = health_data.get("memory_percent", 0)
-        
-        if cpu_percent > 90 or memory_percent > 90:
-            logger.warning(f"High resource usage detected - CPU: {cpu_percent}%, Memory: {memory_percent}%")
-        
-        logger.info("API health check passed successfully")
-
-        return True
-        
-    except requests.exceptions.RequestException as e:
-        logger.error(f"Health check failed: {str(e)}")
-        raise gr.Error("Failed to connect to the API service. Please try again later.")
+            # Check if service is healthy
+            if health_data.get("status") != "healthy":
+                raise gr.Error("API service is not healthy. Please try again later.")
+                
+            # Check resource usage
+            cpu_percent = health_data.get("cpu_percent", 0)
+            memory_percent = health_data.get("memory_percent", 0)
+            
+            if cpu_percent > 90 or memory_percent > 90:
+                logger.warning(f"High resource usage detected - CPU: {cpu_percent}%, Memory: {memory_percent}%")
+            
+            logger.info("API health check passed successfully")
+            return True
+            
+        except requests.exceptions.RequestException as e:
+            last_error = str(e)
+            logger.warning(f"Health check attempt {attempt + 1} failed: {last_error}")
+            if attempt < max_retries - 1:
+                logger.info(f"Waiting {retry_delay} seconds before next attempt...")
+                time.sleep(retry_delay)
+            continue
+    
+    logger.error(f"All health check attempts failed. Last error: {last_error}")
+    raise gr.Error(f"Failed to connect to the API service after {max_retries} attempts. Please try again later.")
 
 def transcribe_youtube(url, return_timestamps, generate_subs, chunk_length_s=15, batch_size=8):
     """Transcribe audio from YouTube video using URL endpoint."""