Fix healthcheck with separate session

lib/backend.py

@@ -190,18 +190,30 @@ class Backend:
             log.debug(f"Exception in main handler loop {e}")
             return web.Response(status=500)
 
+    @cached_property  
+    def healthcheck_session(self):
+        """Dedicated session for healthchecks to avoid conflicts with API session"""
+        log.debug("creating dedicated healthcheck session")
+        connector = TCPConnector(
+            force_close=True,  # Keep this for isolation
+            enable_cleanup_closed=True,
+        )
+        timeout = ClientTimeout(total=10)  # Reasonable timeout for healthchecks
+        return ClientSession(timeout=timeout, connector=connector)
+
     async def __healthcheck(self):
         health_check_url = self.benchmark_handler.healthcheck_endpoint
         if health_check_url is None:
             log.debug("No healthcheck endpoint defined, skipping healthcheck")
             return
+
         while True:
             await sleep(10)
             if self.__start_healthcheck is False:
                 continue
             try:
                 log.debug(f"Performing healthcheck on {health_check_url}")
-                async with self.session.get(health_check_url) as response:
+                async with self.healthcheck_session.get(health_check_url) as response:
                     if response.status == 200:
                         log.debug("Healthcheck successful")
                     elif response.status == 503:
@@ -210,7 +222,6 @@ class Backend:
                             f"Healthcheck failed with status: {response.status}"
                         )
                     else:
-                        # endpoint not ready yet so bail
                         log.debug(f"Healthcheck Endpoint not ready: {response.status}")
             except Exception as e:
                 log.debug(f"Healthcheck failed with exception: {e}")

start_server.sh

@@ -59,12 +59,12 @@ then
     fi
 
     # Fork testing
-    [[ ! -d $SERVER_DIR ]] && git clone "${PYWORKER_REPO:-https://github.com/vast-ai/pyworker}" "$SERVER_DIR"
+    git clone "${PYWORKER_REPO:-https://github.com/vast-ai/pyworker}" "$SERVER_DIR"
     if [[ -n ${PYWORKER_REF:-} ]]; then
         (cd "$SERVER_DIR" && git checkout "$PYWORKER_REF")
     fi
 
-    uv venv --python-preference only-managed "$ENV_PATH" -p 3.10
+    uv venv --managed-python "$ENV_PATH" -p 3.10
     source "$ENV_PATH/bin/activate"
 
     uv pip install -r "${SERVER_DIR}/requirements.txt"

workers/comfyui-json/server.py

@@ -12,7 +12,7 @@ from lib.server import start_server
 from .data_types import ComfyWorkflowData
 
 
-MODEL_SERVER_URL = os.getenv("MODEL_SERVER_URL", "http://127.0.0.1:8288")
+MODEL_SERVER_URL = os.getenv("MODEL_SERVER_URL", "http://127.0.0.1:18288")
 
 # This is the last log line that gets emitted once comfyui+extensions have been fully loaded
 MODEL_SERVER_START_LOG_MSG = "To see the GUI go to: "

workers/comfyui/server.py

@@ -13,10 +13,10 @@ from lib.server import start_server
 from .data_types import DefaultComfyWorkflowData, CustomComfyWorkflowData
 
 
-MODEL_SERVER_URL = "http://127.0.0.1:8288" # API Wrapper Service
+MODEL_SERVER_URL = "http://127.0.0.1:18288" # API Wrapper Service
 
 # This is the last log line that gets emitted once comfyui+extensions have been fully loaded
-MODEL_SERVER_START_LOG_MSG = "To see the GUI go to: http://127.0.0.1:8188"
+MODEL_SERVER_START_LOG_MSG = "To see the GUI go to: http://127.0.0.1:18188"
 MODEL_SERVER_ERROR_LOG_MSGS = [
     "MetadataIncompleteBuffer",  # This error is emitted when the downloaded model is corrupted
     "Value not in list: unet_name",  # This error is emitted when the model file is not there at all