[App] Min replica=0 would break autoscaler component

src/lightning_app/CHANGELOG.md

@@ -51,6 +51,8 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 
 - Fixed a bug where apps that had previously been deleted could not be run again from the CLI ([#16082](https://github.com/Lightning-AI/lightning/pull/16082))
 
+- Fixed a bug where `AutoScaler` would fail with min_replica=0 ([#16092](https://github.com/Lightning-AI/lightning/pull/16092)
+
 
 ## [1.8.4] - 2022-12-08
 

src/lightning_app/components/serve/auto_scaler.py

@@ -224,6 +224,8 @@ def run(self):
         security = HTTPBasic()
         fastapi_app.SEND_TASK = None
 
+        input_type = self._input_type
+
         @fastapi_app.middleware("http")
         async def current_request_counter(request: Request, call_next):
             if not request.scope["path"] == self.endpoint:
@@ -281,7 +283,7 @@ async def update_servers(servers: List[str], authenticated: bool = Depends(authe
             self._iter = cycle(self.servers)
 
         @fastapi_app.post(self.endpoint, response_model=self._output_type)
-        async def balance_api(inputs: self._input_type):
+        async def balance_api(inputs: input_type):
             return await self.process_request(inputs)
 
         endpoint_info_page = self._get_endpoint_info_page()
@@ -578,16 +580,20 @@ def scale(self, replicas: int, metrics: dict) -> int:
             The target number of running works. The value will be adjusted after this method runs
             so that it satisfies ``min_replicas<=replicas<=max_replicas``.
         """
-        pending_requests_per_running_or_pending_work = metrics["pending_requests"] / (
-            replicas + metrics["pending_works"]
-        )
+        pending_requests = metrics["pending_requests"]
+        active_or_pending_works = replicas + metrics["pending_works"]
+
+        if active_or_pending_works == 0:
+            return 1 if pending_requests > 0 else 0
+
+        pending_requests_per_running_or_pending_work = pending_requests / active_or_pending_works
 
-        # scale out if the number of pending requests exceeds max batch size.
+        # scale up if the number of pending requests exceeds max batch size.
         max_requests_per_work = self.max_batch_size
         if pending_requests_per_running_or_pending_work >= max_requests_per_work:
             return replicas + 1
 
-        # scale in if the number of pending requests is below 25% of max_requests_per_work
+        # scale down if the number of pending requests is below 25% of max_requests_per_work
         min_requests_per_work = max_requests_per_work * 0.25
         if pending_requests_per_running_or_pending_work < min_requests_per_work:
             return replicas - 1

tests/tests_app/components/serve/test_auto_scaler.py

@@ -93,6 +93,24 @@ def test_scale(replicas, metrics, expected_replicas):
     assert auto_scaler.scale(replicas, metrics) == expected_replicas
 
 
+def test_scale_from_zero_min_replica():
+    auto_scaler = AutoScaler(
+        EmptyWork,
+        min_replicas=0,
+        max_replicas=2,
+        max_batch_size=10,
+    )
+
+    resp = auto_scaler.scale(0, {"pending_requests": 0, "pending_works": 0})
+    assert resp == 0
+
+    resp = auto_scaler.scale(0, {"pending_requests": 1, "pending_works": 0})
+    assert resp == 1
+
+    resp = auto_scaler.scale(0, {"pending_requests": 1, "pending_works": 1})
+    assert resp <= 0
+
+
 def test_create_work_cloud_compute_cloned():
     """Test CloudCompute is cloned to avoid creating multiple works in a single machine."""
     cloud_compute = CloudCompute("gpu")