From 4938fa79b5c8c7f14c03d9e9200a8689c759ecd6 Mon Sep 17 00:00:00 2001
From: Priyanshu-u07 <connect.priyanshu8271@gmail.com>
Date: Wed, 14 Oct 2026 17:36:39 +0530
Subject: [PATCH] Describe the Kubernetes autoscaler on both auto-replica
 screens

Signed-off-by: Priyanshu-u07 <connect.priyanshu8271@gmail.com>
---
 .../deployment/DeploymentConfig.tsx           |  4 +-
 .../__tests__/autoReplicaConfig.test.tsx      | 39 +++++++++++++++++++
 .../__tests__/autoReplicaSettings.test.tsx    |  7 ++++
 .../components/deployment/configSections.tsx  | 16 +++++++-
 4 files changed, 63 insertions(+), 3 deletions(-)
 create mode 100644 apps/dashboard/src/components/deployment/__tests__/autoReplicaConfig.test.tsx

diff --git a/apps/dashboard/src/components/deployment/DeploymentConfig.tsx b/apps/dashboard/src/components/deployment/DeploymentConfig.tsx
index 53a189d5..015ee575 100644
--- a/apps/dashboard/src/components/deployment/DeploymentConfig.tsx
+++ b/apps/dashboard/src/components/deployment/DeploymentConfig.tsx
@@ -812,7 +812,9 @@ export function AutoReplicaSettings({ autoReplicaEnabled, tokensPerSecondThresho
                         className="w-4 h-4 rounded border-border bg-background text-ember-500 focus:ring-ember-500/40"
                     />
                     <label htmlFor="autoReplicaEnabled" className="text-xs font-bold text-muted-foreground uppercase tracking-tighter">
-                        Automatically provision new nodes when throughput degrades
+                        {isK8s
+                            ? "Scale replicas automatically with request load"
+                            : "Automatically provision new nodes when throughput degrades"}
                     </label>
                 </div>
                 {autoReplicaEnabled && !isK8s && (
diff --git a/apps/dashboard/src/components/deployment/__tests__/autoReplicaConfig.test.tsx b/apps/dashboard/src/components/deployment/__tests__/autoReplicaConfig.test.tsx
new file mode 100644
index 00000000..a42ff2ca
--- /dev/null
+++ b/apps/dashboard/src/components/deployment/__tests__/autoReplicaConfig.test.tsx
@@ -0,0 +1,39 @@
+import { describe, it, expect, vi } from "vitest";
+import { render, screen } from "@testing-library/react";
+
+import { AutoReplicaConfig } from "../configSections";
+
+function renderConfig(provider: string, autoReplicaEnabled = true) {
+  const state = {
+    autoReplicaEnabled,
+    tokensPerSecondThreshold: "10",
+    selectedPool: { provider },
+  };
+  return render(<AutoReplicaConfig state={state as never} dispatch={vi.fn()} />);
+}
+
+describe("the create form describes the autoscaler the pool actually has", () => {
+  it("a kubernetes pool is told about replicas, not nodes", () => {
+    renderConfig("k8s");
+    expect(screen.getByText(/scale replicas automatically/i)).toBeInTheDocument();
+    expect(screen.getByText(/in-flight requests per replica/i)).toBeInTheDocument();
+    expect(screen.queryByText(/provision new nodes/i)).not.toBeInTheDocument();
+  });
+
+  it("a kubernetes pool gets no tokens/sec field, which does nothing there", () => {
+    renderConfig("k8s");
+    expect(screen.queryByLabelText(/tokens\/sec threshold/i)).not.toBeInTheDocument();
+  });
+
+  it("any other pool keeps the node-provisioning text and the tokens/sec field", () => {
+    renderConfig("aws");
+    expect(screen.getByText(/provision new nodes/i)).toBeInTheDocument();
+    expect(screen.getByLabelText(/tokens\/sec threshold/i)).toBeInTheDocument();
+    expect(screen.queryByText(/in-flight requests per replica/i)).not.toBeInTheDocument();
+  });
+
+  it("nothing beyond the toggle shows while auto-replica is off", () => {
+    renderConfig("k8s", false);
+    expect(screen.queryByText(/in-flight requests per replica/i)).not.toBeInTheDocument();
+  });
+});
diff --git a/apps/dashboard/src/components/deployment/__tests__/autoReplicaSettings.test.tsx b/apps/dashboard/src/components/deployment/__tests__/autoReplicaSettings.test.tsx
index 1fdbee24..f5a1ead2 100644
--- a/apps/dashboard/src/components/deployment/__tests__/autoReplicaSettings.test.tsx
+++ b/apps/dashboard/src/components/deployment/__tests__/autoReplicaSettings.test.tsx
@@ -50,6 +50,13 @@ describe("the two autoscalers are never offered together", () => {
       expect(screen.queryByLabelText(label)).toBeNull();
     }
   });
+
+  it("the toggle on a kubernetes pool promises replicas, not new nodes", () => {
+    renderSettings({ isK8s: true });
+
+    expect(screen.getByLabelText(/scale replicas automatically/i)).toBeInTheDocument();
+    expect(screen.queryByText(/provision new nodes/i)).toBeNull();
+  });
 });
 
 describe("when auto-replica is off", () => {
diff --git a/apps/dashboard/src/components/deployment/configSections.tsx b/apps/dashboard/src/components/deployment/configSections.tsx
index cc9333d4..138ff9e6 100644
--- a/apps/dashboard/src/components/deployment/configSections.tsx
+++ b/apps/dashboard/src/components/deployment/configSections.tsx
@@ -238,6 +238,9 @@ export function CompatibilityPanel({
 
 /** Auto-replica scale-out config (inference deploys). */
 export function AutoReplicaConfig({ state, dispatch }: SectionProps) {
+  // Kubernetes scales replicas through KEDA; the tokens/sec threshold drives
+  // node provisioning, which only the other pool types do.
+  const isK8s = state.selectedPool?.provider === "k8s"
   return (
     <div className="space-y-4 p-4 bg-muted/50 rounded-lg border">
       <div className="flex items-center gap-2 mb-2">
@@ -253,10 +256,19 @@ export function AutoReplicaConfig({ state, dispatch }: SectionProps) {
           className="w-4 h-4 rounded border-border"
         />
         <label htmlFor="autoReplicaEnabled" className="text-xs font-medium text-muted-foreground">
-          Automatically provision new nodes when throughput degrades
+          {isK8s
+            ? "Scale replicas automatically with request load"
+            : "Automatically provision new nodes when throughput degrades"}
         </label>
       </div>
-      {state.autoReplicaEnabled && (
+      {state.autoReplicaEnabled && isK8s && (
+        <p className="text-xs text-muted-foreground">
+          Replicas scale on in-flight requests per replica, with p95 latency as a
+          guardrail. Platform defaults apply until you change the targets in the
+          deployment's Configuration tab.
+        </p>
+      )}
+      {state.autoReplicaEnabled && !isK8s && (
         <div>
           <label htmlFor="tokensPerSecondThreshold" className="block text-xs font-medium text-muted-foreground mb-1.5">
             Tokens/sec threshold <span className="text-muted-foreground/60">(scale out when average drops below)</span>