feat: Add exponential backoff for session lock retries in ClickHouse client

src/db/db_connection.py

@@ -18,26 +18,50 @@ def create_client():
     if not clickhouse_password:
         raise ValueError("CLICKHOUSE_PASSWORD environment variable not set.")
 
-    max_retries = 3
+    max_retries = 5  # Increased from 3 to 5
-    retry_delay = 2  # seconds
+    retry_delay = 1  # Reduced initial delay to 1 second
-
+    
+    last_exception = None
+    
     for attempt in range(max_retries):
         try:
+            # Generate a new UUID for each attempt
+            session_id = str(uuid.uuid4())
+            
             client = clickhouse_connect.get_client(
                 host="clickhouse.abellana.work",
                 port=443,
                 username="default",
                 password=clickhouse_password,
                 secure=True,
-                session_id=str(uuid.uuid4()),
+                session_id=session_id,
-                settings={'session_timeout': 60}
+                settings={
+                    'session_timeout': 60,
+                    'session_check': 1
+                }
             )
-            # Test the connection.
+            
+            # Test the connection
             client.query('SELECT 1')
             return client
+            
         except Exception as e:
-            if attempt < max_retries - 1:
+            last_exception = e
+            
+            # Check specifically for session locked error
+            if "SESSION_IS_LOCKED" in str(e):
+                if attempt < max_retries - 1:
+                    # Use exponential backoff for session locks
+                    wait_time = retry_delay * (2 ** attempt)
+                    logging.warning(f"Session locked, retrying in {wait_time} seconds...")
+                    time.sleep(wait_time)
+                    continue
+            
+            # For other errors, use normal retry logic
+            elif attempt < max_retries - 1:
                 logging.warning(f"Connection attempt {attempt + 1} failed: {str(e)}")
                 time.sleep(retry_delay)
-            else:
+                continue
-                raise
+            
+            # If we've exhausted all retries, raise the last exception
+            raise last_exception