fix health check issue

Author: shashikg

.gitignore

@@ -132,3 +132,6 @@ dmypy.json
 
 # Pyre type checker
 .pyre/
+
+# Mac DS_Store
+.DS_Store

fastdeploy/_app.py

@@ -303,23 +303,27 @@ def on_get(self, req, resp):
 class Health(object):
     def on_get(self, req, resp):
         stuck_for = float(req.params.get("stuck"))
-        predictor_wait_started_at = _utils.META_INDEX["predictor_wait_started_at"]
-
-        if (
-            stuck_for
-            and predictor_wait_started_at
-            and time.time() - predictor_wait_started_at >= stuck_for
-        ):
-            resp.status = falcon.HTTP_503
-            resp.media = {
-                "predictor_status": f"predictor stuck for {time.time() - predictor_wait_started_at}. deemed stuck."
-            }
+        last_prediction_loop_start_time = _utils.META_INDEX["last_prediction_loop_start_time"]
+        prediction_loop_stuck_for = time.time() - last_prediction_loop_start_time
+
+        if last_prediction_loop_start_time:
+            if (
+                stuck_for
+                and prediction_loop_stuck_for >= stuck_for
+            ):
+                resp.status = falcon.HTTP_503
+                resp.media = {
+                    "predictor_status": f"prediction loop stuck for {prediction_loop_stuck_for}. deemed stuck."
+                }
+            else:
+                resp.status = falcon.HTTP_200
+                resp.media = {
+                    "predictor_status": f"prediction loop running for {prediction_loop_stuck_for}"
+                }
         else:
             resp.status = falcon.HTTP_200
             resp.media = {
-                "predictor_status": f"queue waiting for predictor for {time.time() - predictor_wait_started_at}"
-                if predictor_wait_started_at
-                else "queue empty"
+                "predictor_status": f"prediction loop not started"
             }
 
 

fastdeploy/_loop.py

@@ -10,6 +10,8 @@
 
 
 def start_loop():
+    _utils.META_INDEX["last_prediction_loop_start_time"] = 0
+
     ACCEPTS_EXTRAS = False
     try:
         from predictor import predictor
@@ -64,7 +66,8 @@ def start_loop():
         first_sleep_start_time = 0
         __loop_is_sleeping = False
         while True:
-            _utils.META_INDEX["predictor_wait_started_at"] = 0
+            _utils.META_INDEX["last_prediction_loop_start_time"] = time.time()
+            
             if len(_utils.REQUEST_INDEX):
                 (
                     unique_id,
@@ -125,7 +128,6 @@ def start_loop():
         try:
             pred_start_time = time.time()
             __in_batch_length = len(batch)
-            _utils.META_INDEX["predictor_wait_started_at"] = time.time()
 
             if ACCEPTS_EXTRAS:
                 preds = predictor(
@@ -134,8 +136,6 @@ def start_loop():
             else:
                 preds = predictor(batch, batch_size=batch_size)
 
-            _utils.META_INDEX["predictor_wait_started_at"] = 0
-
             if not isinstance(preds, list) or len(preds) != __in_batch_length:
                 _utils.logger.error(
                     "Something is seriously wrong! len(inputs) != len(outputs) from predictor.py. Check your recipe"