[dashboard] Pipe resource assignments to dashboard (#8998)

python/ray/dashboard/client/src/api.ts

@@ -133,9 +133,18 @@ export type NodeInfoResponse = {
 
 export const getNodeInfo = () => get<NodeInfoResponse>("/api/node_info", {});
 
+export type ResourceSlot = {
+  slot: number;
+  allocation: number;
+};
+
+export type ResourceAllocations = {
+  resourceSlots: ResourceSlot[];
+};
+
 export type RayletCoreWorkerStats = {
   usedResources: {
-    [key: string]: number;
+    [key: string]: ResourceAllocations;
   };
 };
 
@@ -168,7 +177,7 @@ export type RayletActorInfo =
       taskQueueLength: number;
       timestamp: number;
       usedObjectStoreMemory: number;
-      usedResources: { [key: string]: number };
+      usedResources: { [key: string]: ResourceAllocations };
       currentTaskDesc?: string;
       numPendingTasks?: number;
       webuiDisplay?: Record<string, string>;

python/ray/dashboard/client/src/pages/dashboard/logical-view/Actor.tsx

@@ -16,6 +16,7 @@ import {
   launchProfiling,
   RayletActorInfo,
 } from "../../../api";
+import { sum } from "../../../common/util";
 import ActorDetailsPane from "./ActorDetailsPane";
 import Actors from "./Actors";
 
@@ -137,7 +138,12 @@ class Actor extends React.Component<Props & WithStyles<typeof styles>, State> {
                 Object.entries(actor.usedResources).length > 0 &&
                 Object.entries(actor.usedResources)
                   .sort((a, b) => a[0].localeCompare(b[0]))
-                  .map(([key, value]) => `${value.toLocaleString()} ${key}`)
+                  .map(
+                    ([key, value]) =>
+                      `${sum(
+                        value.resourceSlots.map((slot) => slot.allocation),
+                      )} ${key}`,
+                  )
                   .join(", "),
             },
             {

python/ray/dashboard/client/src/pages/dashboard/node-info/features/GPU.tsx

@@ -63,13 +63,22 @@ export const NodeGPU: NodeFeatureComponent = ({ node }) => {
 
 export const WorkerGPU: WorkerFeatureComponent = ({ rayletWorker }) => {
   const workerRes = rayletWorker?.coreWorkerStats.usedResources;
-  const workerUsedGPUResources = workerRes?.["GPU"] || NaN;
-  const message = isNaN(workerUsedGPUResources) ? (
-    <Typography color="textSecondary" component="span" variant="inherit">
-      N/A
-    </Typography>
-  ) : (
-    <b>`${workerUsedGPUResources} GPUs in use`</b>
-  );
+  const workerUsedGPUResources = workerRes?.["GPU"];
+  let message;
+  if (workerUsedGPUResources === undefined) {
+    message = (
+      <Typography color="textSecondary" component="span" variant="inherit">
+        N/A
+      </Typography>
+    );
+  } else {
+    const aggregateAllocation = sum(
+      workerUsedGPUResources.resourceSlots.map(
+        (resourceSlot) => resourceSlot.allocation,
+      ),
+    );
+    const plural = aggregateAllocation === 1 ? "" : "s";
+    message = <b>{`${aggregateAllocation} GPU${plural} in use`}</b>;
+  }
   return <div style={{ minWidth: 60 }}>{message}</div>;
 };

python/ray/tests/test_metrics.py

@@ -223,7 +223,13 @@ def test_raylet_info_endpoint(shutdown_only):
                 raise Exception(
                     "Timed out while waiting for dashboard to start.")
 
-    assert parent_actor_info["usedResources"]["CPU"] == 2
+    def cpu_resources(actor_info):
+        cpu_resources = 0
+        for slot in actor_info["usedResources"]["CPU"]["resourceSlots"]:
+            cpu_resources += slot["allocation"]
+        return cpu_resources
+
+    assert cpu_resources(parent_actor_info) == 2
     assert parent_actor_info["numExecutedTasks"] == 4
     for _, child_actor_info in children.items():
         if child_actor_info["state"] == -1:
@@ -231,7 +237,7 @@ def test_raylet_info_endpoint(shutdown_only):
         else:
             assert child_actor_info["state"] == 1
             assert len(child_actor_info["children"]) == 0
-            assert child_actor_info["usedResources"]["CPU"] == 1
+            assert cpu_resources(child_actor_info) == 1
 
     profiling_id = requests.get(
         webui_url + "/api/launch_profiling",