From eb7da0a2e558456bdb00ec3cf6cfb00ad3eaf40a Mon Sep 17 00:00:00 2001
From: Vincent Koc <vincentkoc@ieee.org>
Date: Fri, 19 Jun 2026 12:03:31 +0200
Subject: [PATCH] fix(plugins): cancel self-hosted probe error bodies

---
 .../provider-self-hosted-setup.test.ts        | 44 ++++++++++++++++++-
 src/plugins/provider-self-hosted-setup.ts     |  8 ++++
 2 files changed, 51 insertions(+), 1 deletion(-)

diff --git a/src/plugins/provider-self-hosted-setup.test.ts b/src/plugins/provider-self-hosted-setup.test.ts
index 9f2f72ed2ac..d6429992438 100644
--- a/src/plugins/provider-self-hosted-setup.test.ts
+++ b/src/plugins/provider-self-hosted-setup.test.ts
@@ -86,10 +86,30 @@ async function configureSelfHostedTestProvider(params: {
   });
 }
 
+function cancelTrackedResponse(init?: ResponseInit): {
+  response: Response;
+  wasCanceled: () => boolean;
+} {
+  let canceled = false;
+  const stream = new ReadableStream<Uint8Array>({
+    start(controller) {
+      controller.enqueue(new TextEncoder().encode("ignored"));
+    },
+    cancel() {
+      canceled = true;
+    },
+  });
+  return {
+    response: new Response(stream, init),
+    wasCanceled: () => canceled,
+  };
+}
+
 describe("discoverOpenAICompatibleLocalModels", () => {
   it("uses guarded fetch pinned to the configured self-hosted provider", async () => {
     const release = vi.fn(async () => undefined);
     const propsRelease = vi.fn(async () => undefined);
+    const propsResponse = cancelTrackedResponse({ status: 404 });
     fetchWithSsrFGuardMock.mockResolvedValueOnce({
       response: new Response(JSON.stringify({ data: [{ id: "Qwen/Qwen3-32B" }] }), {
         status: 200,
@@ -98,7 +118,7 @@ describe("discoverOpenAICompatibleLocalModels", () => {
       release,
     });
     fetchWithSsrFGuardMock.mockResolvedValueOnce({
-      response: new Response("{}", { status: 404 }),
+      response: propsResponse.response,
       finalUrl: "http://127.0.0.1:8000/props",
       release: propsRelease,
     });
@@ -141,6 +161,28 @@ describe("discoverOpenAICompatibleLocalModels", () => {
     });
     expect(release).toHaveBeenCalledOnce();
     expect(propsRelease).toHaveBeenCalledOnce();
+    expect(propsResponse.wasCanceled()).toBe(true);
+  });
+
+  it("cancels model discovery error bodies before falling back", async () => {
+    const release = vi.fn(async () => undefined);
+    const response = cancelTrackedResponse({ status: 503, statusText: "Service Unavailable" });
+    fetchWithSsrFGuardMock.mockResolvedValueOnce({
+      response: response.response,
+      finalUrl: "http://127.0.0.1:8000/v1/models",
+      release,
+    });
+
+    const models = await discoverOpenAICompatibleLocalModels({
+      baseUrl: "http://127.0.0.1:8000/v1/",
+      apiKey: "self-hosted-test-key",
+      label: "vLLM",
+      env: {},
+    });
+
+    expect(models).toEqual([]);
+    expect(release).toHaveBeenCalledOnce();
+    expect(response.wasCanceled()).toBe(true);
   });
 
   it("uses llama.cpp nested /props n_ctx as the runtime context cap", async () => {
diff --git a/src/plugins/provider-self-hosted-setup.ts b/src/plugins/provider-self-hosted-setup.ts
index 662c4c3ca43..f66f8ef9fea 100644
--- a/src/plugins/provider-self-hosted-setup.ts
+++ b/src/plugins/provider-self-hosted-setup.ts
@@ -86,6 +86,12 @@ function readPositiveInteger(value: unknown): number | undefined {
   return Math.trunc(value);
 }
 
+async function cancelUnreadResponseBody(response: Response): Promise<void> {
+  if (!response.bodyUsed) {
+    await response.body?.cancel().catch(() => undefined);
+  }
+}
+
 function resolveLlamaCppPropsUrl(baseUrl: string, modelId?: string): string {
   const parsed = new URL(baseUrl);
   const pathname = parsed.pathname.replace(/\/+$/, "");
@@ -124,6 +130,7 @@ async function discoverLlamaCppRuntimeContextTokens(params: {
     });
     try {
       if (!response.ok) {
+        await cancelUnreadResponseBody(response);
         return undefined;
       }
       const data = (await response.json()) as LlamaCppPropsResponse;
@@ -167,6 +174,7 @@ export async function discoverOpenAICompatibleLocalModels(params: {
     });
     try {
       if (!response.ok) {
+        await cancelUnreadResponseBody(response);
         log.warn(`Failed to discover ${params.label} models: ${response.status}`);
         return [];
       }