Gate load shedding contribution rule

Author: corlettb

app/config.py

@@ -168,6 +168,10 @@ class Config:
     HIGH_WATER_MARK = int(os.getenv("HIGH_WATER_MARK", "26"))
     # Throttle services contributing this % or more of total request volume
     THROTTLE_CONTRIBUTION_PCT = int(os.getenv("THROTTLE_CONTRIBUTION_PCT", "20"))
+    # Only apply contribution-based throttling when enough services are active
+    THROTTLE_CONTRIBUTION_MIN_SERVICES = int(os.getenv("THROTTLE_CONTRIBUTION_MIN_SERVICES", "5"))
+    # Only apply contribution-based throttling when total volume is above this threshold
+    THROTTLE_CONTRIBUTION_MIN_VOLUME = int(os.getenv("THROTTLE_CONTRIBUTION_MIN_VOLUME", "50"))
     # Throttle services with volume this many times above median
     THROTTLE_VOLUME_MEDIAN_MULTIPLE = int(os.getenv("THROTTLE_VOLUME_MEDIAN_MULTIPLE", "10"))
 

app/load_shedding.py

@@ -254,15 +254,19 @@ def should_throttle_service(service_id: str) -> bool:
 
         # Check if contributing a significant % of load
         throttle_contribution_pct = current_app.config.get("THROTTLE_CONTRIBUTION_PCT", 20)
-        if contribution_pct >= throttle_contribution_pct:
-            current_app.logger.info(
-                "Service %s contributing %.1f%% of load (%s/%s requests)",
-                service_id,
-                contribution_pct,
-                current_volume,
-                total_volume,
-            )
-            return True
+        min_services = current_app.config.get("THROTTLE_CONTRIBUTION_MIN_SERVICES", 5)
+        min_total_volume = current_app.config.get("THROTTLE_CONTRIBUTION_MIN_VOLUME", 50)
+
+        if len(service_volumes) >= min_services and total_volume >= min_total_volume:
+            if contribution_pct >= throttle_contribution_pct:
+                current_app.logger.info(
+                    "Service %s contributing %.1f%% of load (%s/%s requests)",
+                    service_id,
+                    contribution_pct,
+                    current_volume,
+                    total_volume,
+                )
+                return True
 
         # Check if volume is significantly above median (outlier detection)
         volumes = sorted(service_volumes.values())

tests/app/test_load_shedding.py

@@ -204,6 +204,8 @@ def test_throttles_service_with_volume_above_median_multiple(
 
     def test_single_service_contributing_100_pct(self, notify_api: "Flask", mocker: MockerFixture) -> None:
         mocker.patch("app.load_shedding.is_worker_overloaded", return_value=True)
+        notify_api.config["THROTTLE_CONTRIBUTION_MIN_SERVICES"] = 1
+        notify_api.config["THROTTLE_CONTRIBUTION_MIN_VOLUME"] = 1
 
         # Test with single service (contributes 100% of load)
         # Should be throttled by contribution % (100% > 20% threshold)
@@ -216,6 +218,38 @@ def test_single_service_contributing_100_pct(self, notify_api: "Flask", mocker:
         # Single service contributing 100% should be throttled (exceeds 20% threshold)
         assert should_throttle_service("single") is True
 
+    def test_does_not_throttle_by_contribution_with_few_services(
+        self, notify_api: "Flask", mocker: MockerFixture
+    ) -> None:
+        mocker.patch("app.load_shedding.is_worker_overloaded", return_value=True)
+        notify_api.config["THROTTLE_CONTRIBUTION_MIN_SERVICES"] = 5
+        notify_api.config["THROTTLE_CONTRIBUTION_MIN_VOLUME"] = 50
+
+        # Only 2 services, high contribution but below min service count
+        service_volumes = {"big": 9, "small": 1}
+
+        mock_tracker = mocker.patch("app.load_shedding._volume_tracker")
+        mock_tracker.get_volumes.return_value = service_volumes
+
+        # Should not throttle by contribution; median multiple also not met
+        assert should_throttle_service("big") is False
+
+    def test_does_not_throttle_by_contribution_with_low_total_volume(
+        self, notify_api: "Flask", mocker: MockerFixture
+    ) -> None:
+        mocker.patch("app.load_shedding.is_worker_overloaded", return_value=True)
+        notify_api.config["THROTTLE_CONTRIBUTION_MIN_SERVICES"] = 3
+        notify_api.config["THROTTLE_CONTRIBUTION_MIN_VOLUME"] = 50
+
+        # 3 services but total volume below min threshold
+        service_volumes = {"big": 8, "mid": 1, "small": 1}
+
+        mock_tracker = mocker.patch("app.load_shedding._volume_tracker")
+        mock_tracker.get_volumes.return_value = service_volumes
+
+        # Should not throttle by contribution; median multiple also not met
+        assert should_throttle_service("big") is False
+
     def test_returns_false_on_error(self, notify_api: "Flask", mocker: MockerFixture) -> None:
         mocker.patch("app.load_shedding.is_worker_overloaded", return_value=True)