[core][dashboard] Return 4xx from node and actor detail APIs

python/ray/dashboard/client/src/pages/actor/hook/useActorDetail.ts

@@ -1,3 +1,4 @@
+import axios from "axios";
 import { useContext, useState } from "react";
 import { useParams } from "react-router-dom";
 import useSWR from "swr";
@@ -9,12 +10,20 @@ export const useFetchActor = (actorId: string | null) => {
   return useSWR(
     actorId ? ["useActorDetail", actorId] : null,
     async ([_, actorId]) => {
-      const actor_resp = await getActor(actorId);
-      const data: ActorResp = actor_resp?.data;
-      const { data: rspData } = data;
+      try {
+        const actor_resp = await getActor(actorId);
+        const data: ActorResp = actor_resp?.data;
+        const { data: rspData } = data;
 
-      if (rspData.detail) {
-        return rspData.detail;
+        if (rspData.detail) {
+          return rspData.detail;
+        }
+      } catch (e) {
+        // The API returns 404 for an unknown actor ID, which axios rejects on.
+        if (axios.isAxiosError(e) && e.response?.status === 404) {
+          return undefined;
+        }
+        throw e;
       }
     },
   );
@@ -28,19 +37,24 @@ export const useActorDetail = () => {
   const { data: actorDetail, isLoading } = useSWR(
     ["useActorDetail", params.actorId],
     async ([_, actorId]) => {
-      const actor_resp = await getActor(actorId);
-      const data: ActorResp = actor_resp?.data;
-      const { data: rspData, msg, result } = data;
-      if (msg) {
-        setMsg(msg);
-      }
-
-      if (result === false) {
-        setMsg("Actor Query Error Please Check Actor Id");
-      }
+      try {
+        const actor_resp = await getActor(actorId);
+        const data: ActorResp = actor_resp?.data;
+        const { data: rspData, msg } = data;
+        if (msg) {
+          setMsg(msg);
+        }
 
-      if (rspData.detail) {
-        return rspData.detail;
+        if (rspData.detail) {
+          return rspData.detail;
+        }
+      } catch (e) {
+        // The API returns 404 for an unknown actor ID, which axios rejects on.
+        if (axios.isAxiosError(e) && e.response?.status === 404) {
+          setMsg("Actor Query Error Please Check Actor Id");
+          return undefined;
+        }
+        throw e;
       }
     },
     { refreshInterval: API_REFRESH_INTERVAL_MS },

python/ray/dashboard/client/src/pages/node/hook/useNodeDetail.ts

@@ -1,3 +1,4 @@
+import axios from "axios";
 import { useContext, useState } from "react";
 import { useParams } from "react-router-dom";
 import useSWR from "swr";
@@ -18,20 +19,27 @@ export const useNodeDetail = () => {
   const { data: nodeDetail, isLoading } = useSWR(
     ["useNodeDetail", params.id],
     async ([_, nodeId]) => {
-      const { data } = await getNodeDetail(nodeId);
-      const { data: rspData, msg, result } = data;
+      try {
+        const { data } = await getNodeDetail(nodeId);
+        const { data: rspData, msg } = data;
 
-      if (msg) {
-        setMsg(msg);
-      }
-
-      if (result === false) {
-        setMsg("Node Query Error Please Check Node Name");
-        setRefresh(false);
-      }
+        if (msg) {
+          setMsg(msg);
+        }
 
-      if (rspData?.detail) {
-        return rspData.detail;
+        if (rspData?.detail) {
+          return rspData.detail;
+        }
+      } catch (e) {
+        // The API returns 404 for an unknown node ID, which axios rejects on.
+        // Keep auto-refresh running: a 404 can be transient (e.g. the node is
+        // not in the dashboard's node table yet), and stopping the refresh
+        // would leave the page stuck on the error until a manual reload.
+        if (axios.isAxiosError(e) && e.response?.status === 404) {
+          setMsg("Node Query Error Please Check Node Name");
+          return undefined;
+        }
+        throw e;
       }
     },
     { refreshInterval: isRefreshing ? API_REFRESH_INTERVAL_MS : 0 },

python/ray/dashboard/modules/node/datacenter.py

@@ -127,9 +127,12 @@ def _extract_workers_for_node(cls, node_physical_stats, node_stats):
 
     @classmethod
     async def get_node_info(cls, node_id, get_summary=False):
+        node = DataSource.nodes.get(node_id)
+        if node is None:
+            return None
+
         node_physical_stats = dict(DataSource.node_physical_stats.get(node_id, {}))
         node_stats = dict(DataSource.node_stats.get(node_id, {}))
-        node = DataSource.nodes.get(node_id, {})
 
         if get_summary:
             node_physical_stats.pop("workers", None)

python/ray/dashboard/modules/node/node_head.py

@@ -413,7 +413,7 @@ async def get_all_nodes(self, req) -> aiohttp.web.Response:
             )
         else:
             return dashboard_optional_utils.rest_response(
-                status_code=dashboard_utils.HTTPStatusCode.INTERNAL_ERROR,
+                status_code=dashboard_utils.HTTPStatusCode.BAD_REQUEST,
                 message=f"Unknown view {view}",
             )
 
@@ -422,6 +422,11 @@ async def get_all_nodes(self, req) -> aiohttp.web.Response:
     async def get_node(self, req) -> aiohttp.web.Response:
         node_id = req.match_info.get("node_id")
         node_info = await DataOrganizer.get_node_info(node_id)
+        if node_info is None:
+            return dashboard_optional_utils.rest_response(
+                status_code=dashboard_utils.HTTPStatusCode.NOT_FOUND,
+                message=f"Node {node_id} not found.",
+            )
         return dashboard_optional_utils.rest_response(
             status_code=dashboard_utils.HTTPStatusCode.OK,
             message="Node details fetched.",
@@ -718,10 +723,17 @@ async def get_all_actors(self, req) -> aiohttp.web.Response:
     async def get_actor(self, req) -> aiohttp.web.Response:
         actor_id = req.match_info.get("actor_id")
         actors = await DataOrganizer.get_actor_infos(actor_ids=[actor_id])
+        # `get_actor_infos` returns a `None` entry for unknown actor IDs.
+        actor_detail = actors.get(actor_id)
+        if actor_detail is None:
+            return dashboard_optional_utils.rest_response(
+                status_code=dashboard_utils.HTTPStatusCode.NOT_FOUND,
+                message=f"Actor {actor_id} not found.",
+            )
         return dashboard_optional_utils.rest_response(
             status_code=dashboard_utils.HTTPStatusCode.OK,
             message="Actor details fetched.",
-            detail=actors[actor_id],
+            detail=actor_detail,
         )
 
     @routes.get("/test/dump")

python/ray/dashboard/modules/node/tests/test_actor.py

@@ -374,5 +374,17 @@ class InfeasibleActor:
                 raise Exception(f"Timed out while testing, {ex_stack}")
 
 
+def test_actor_not_found_status_code(disable_aiohttp_cache, ray_start_with_dashboard):
+    """`GET /logical/actors/{actor_id}` returns 404 for an unknown actor ID."""
+    assert wait_until_server_available(ray_start_with_dashboard["webui_url"]) is True
+    webui_url = format_web_url(ray_start_with_dashboard["webui_url"])
+
+    resp = requests.get(f"{webui_url}/logical/actors/{'f' * 32}")
+    assert resp.status_code == 404, resp.text
+    resp_json = resp.json()
+    assert resp_json["result"] is False
+    assert "not found" in resp_json["msg"]
+
+
 if __name__ == "__main__":
     sys.exit(pytest.main(["-v", __file__]))

python/ray/dashboard/modules/node/tests/test_node.py

@@ -412,5 +412,51 @@ def _check_worker_pids():
     wait_for_condition(_check_worker_pids, timeout=20)
 
 
+def _wait_for_node_registered(webui_url, node_id, timeout=30):
+    """Wait until `node_id` appears in the dashboard's node summary."""
+
+    def _registered():
+        response = requests.get(webui_url + "/nodes?view=summary")
+        if response.status_code != 200:
+            return False
+        summary = response.json()["data"]["summary"]
+        return any(node["raylet"]["nodeId"] == node_id for node in summary)
+
+    wait_for_condition(_registered, timeout=timeout)
+
+
+def test_node_api_status_codes(disable_aiohttp_cache, ray_start_with_dashboard):
+    """The node APIs return 4xx for client errors instead of 200/500.
+
+    `ray_start_with_dashboard` spins up a fresh Ray cluster, so this is a
+    single test that queries one cluster multiple times rather than a
+    parametrized test (which would launch a cluster per case).
+    """
+    assert wait_until_server_available(ray_start_with_dashboard["webui_url"]) is True
+    webui_url = format_web_url(ray_start_with_dashboard["webui_url"])
+    node_id = ray_start_with_dashboard["node_id"]
+
+    # `wait_until_server_available` only checks that the HTTP server accepts
+    # connections. `/nodes/{node_id}` now 404s until the node shows up in the
+    # dashboard's node table, so wait for that before asserting on a status code.
+    _wait_for_node_registered(webui_url, node_id)
+
+    unknown_node_id = "8" * len(node_id)
+    for path, expected_status_code in [
+        # An unknown node ID is a client error, not a server error.
+        (f"/nodes/{unknown_node_id}", 404),
+        # An unsupported `view` is a client error, not a server error.
+        ("/nodes?view=unknown_view", 400),
+        ("/nodes", 400),
+        # Sanity check that the success paths still return 200.
+        (f"/nodes/{node_id}", 200),
+        ("/nodes?view=summary", 200),
+        ("/nodes?view=hostnamelist", 200),
+    ]:
+        response = requests.get(webui_url + path)
+        assert response.status_code == expected_status_code, (path, response.text)
+        assert response.json()["result"] is (expected_status_code == 200), path
+
+
 if __name__ == "__main__":
     sys.exit(pytest.main(["-v", __file__]))