perf(clicks): omit None fields from time-series inserts

Explicit nulls flip a bucket field's BSON type and force the bucket
to close early, fragmenting the collection into a few measurements
per bucket. Absent fields pack fine and queries are unaffected:
MongoDB treats missing like null in $eq/$group and the stats
sentinels already map both.

Author: Zingzy

schemas/models/click.py

@@ -60,3 +60,14 @@ class ClickDoc(MongoBaseModel):
     utm_source: str | None = None
     utm_medium: str | None = None
     utm_campaign: str | None = None
+
+    def to_mongo(self) -> dict:
+        """Serialise for insertion, omitting None fields entirely.
+
+        Time-series buckets track one BSON type per field. An explicit
+        null following a string value (or vice versa) is a type conflict
+        that closes the bucket early, while an absent field packs fine.
+        Queries are unaffected: MongoDB treats missing and null the same
+        in $eq/$group, and the stats sentinels already map both.
+        """
+        return self.model_dump(by_alias=True, exclude_none=True)

tests/unit/schemas/models/test_click.py

@@ -50,3 +50,19 @@ def test_to_mongo_round_trip(self):
         restored = ClickDoc.from_mongo(doc.to_mongo())
         assert restored.referrer == "google.com"
         assert restored.redirect_ms == doc.redirect_ms
+
+    def test_to_mongo_omits_none_fields(self):
+        # Explicit nulls flip the per-field BSON type and close time-series
+        # buckets; None fields must be absent from the insert document.
+        data = self._make(referrer=None, bot_name=None).to_mongo()
+        assert "referrer" not in data
+        assert "bot_name" not in data
+        assert "device" not in data
+        assert "utm_source" not in data
+        assert "domain" not in data["meta"]
+
+    def test_to_mongo_keeps_set_fields(self):
+        data = self._make(referrer="google.com", device="mobile").to_mongo()
+        assert data["referrer"] == "google.com"
+        assert data["device"] == "mobile"
+        assert data["country"] == "Unknown"

tests/unit/services/test_click_service.py

@@ -303,10 +303,11 @@ async def test_utm_tags_default_to_none(self):
 
         await handler.handle(make_context(url_data))
 
+        # Absent, not null: explicit nulls close time-series buckets.
         doc = d.click_repo.insert.call_args[0][0]
-        assert doc["utm_source"] is None
-        assert doc["utm_medium"] is None
-        assert doc["utm_campaign"] is None
+        assert "utm_source" not in doc
+        assert "utm_medium" not in doc
+        assert "utm_campaign" not in doc
 
     @pytest.mark.asyncio
     async def test_meta_carries_domain_from_cache(self):
@@ -320,17 +321,18 @@ async def test_meta_carries_domain_from_cache(self):
         assert doc["meta"]["domain"] == "links.acme.com"
 
     @pytest.mark.asyncio
-    async def test_meta_domain_none_when_cache_empty_string(self):
-        # Older cached entries pre-PR1 have domain="". Coerce to None so
-        # per-domain queries can distinguish "unknown" from a real value.
+    async def test_meta_domain_omitted_when_cache_empty_string(self):
+        # Older cached entries pre-PR1 have domain="". Coerced to None and
+        # then omitted from the insert doc ($eq: null matches missing, so
+        # per-domain queries still distinguish "unknown" from a real value).
         d = make_deps()
         handler = make_v2_handler(d.click_repo, d.url_repo, d.geoip, d.url_cache)
         url_data = make_v2_cache(domain="")
 
         await handler.handle(make_context(url_data))
 
         doc = d.click_repo.insert.call_args[0][0]
-        assert doc["meta"]["domain"] is None
+        assert "domain" not in doc["meta"]
 
     @pytest.mark.asyncio
     async def test_blocked_bot_skips_analytics_no_error(self):