Fix nan values (#2665)

* Fix nan values

* unused imports

ee/clickhouse/models/event.py

@@ -112,7 +112,9 @@ def get_properties(self, event):
             prop_vals = [res.strip('"') for res in event[9]]
             return dict(zip(event[8], prop_vals))
         else:
-            props = json.loads(event[2])
+            # parse_constants gets called for any NaN, Infinity etc values
+            # we just want those to be returned as None
+            props = json.loads(event[2], parse_constant=lambda x: None)
             unpadded = {key: value.strip('"') if isinstance(value, str) else value for key, value in props.items()}
             return unpadded
 

posthog/api/test/test_capture.py

@@ -7,6 +7,7 @@
 from urllib.parse import quote
 
 import lzstring
+import numpy as np
 from django.utils import timezone
 from freezegun import freeze_time
 
@@ -495,3 +496,15 @@ def test_incorrect_json(self):
             "/capture/", '{"event": "incorrect json with trailing comma",}', content_type="application/json"
         )
         self.assertEqual(response.json()["code"], "validation")
+
+    @patch("posthog.api.capture.celery_app.send_task")
+    def test_nan(self, patch_process_event_with_plugins):
+        self.client.post(
+            "/track/",
+            data={
+                "data": json.dumps([{"event": "beep", "properties": {"distinct_id": np.nan}}]),
+                "api_key": self.team.api_token,
+            },
+        )
+        arguments = self._to_arguments(patch_process_event_with_plugins)
+        self.assertEqual(arguments["data"]["properties"]["distinct_id"], None)

posthog/utils.py

@@ -342,7 +342,10 @@ def load_data_from_request(request):
 
     #  Is it plain json?
     try:
-        data = json.loads(data)
+        # parse_constant gets called in case of NaN, Infinity etc
+        # default behaviour is to put those into the DB directly
+        # but we just want it to return None
+        data = json.loads(data, parse_constant=lambda x: None)
     except json.JSONDecodeError:
         # if not, it's probably base64 encoded from other libraries
         data = base64_to_json(data)