Add E2E regression test for GET/HEAD requests with a body (#458)

app/v3/helloWorld/function.json

@@ -5,7 +5,7 @@
             "type": "httpTrigger",
             "direction": "in",
             "name": "req",
-            "methods": ["get", "post"]
+            "methods": ["get", "post", "head"]
         },
         {
             "type": "http",

app/v4/src/functions/helloWorld.ts

@@ -12,7 +12,7 @@ export async function helloWorld(request: HttpRequest, context: InvocationContex
 }
 
 app.http('helloWorld', {
-    methods: ['GET', 'POST'],
+    methods: ['GET', 'POST', 'HEAD'],
     authLevel: 'anonymous',
     handler: helloWorld,
 });

src/http.test.ts

@@ -3,6 +3,7 @@
 
 import Agent from 'agentkeepalive';
 import { expect } from 'chai';
+import * as http from 'http';
 import { encode } from 'iconv-lite';
 // Node.js core added support for fetch in v18, but while we're testing versions <18 we'll use "node-fetch"
 import { default as fetch, HeadersInit } from 'node-fetch';
@@ -26,6 +27,42 @@ const octetStreamHeaders = getContentTypeHeaders('application/octet-stream');
 const multipartFormHeaders = getContentTypeHeaders('multipart/form');
 const textPlainHeaders = getContentTypeHeaders('text/plain');
 
+// node-fetch (like the WHATWG fetch spec) refuses to send a body with GET/HEAD, so we use the raw
+// http module to reproduce a GET/HEAD request that arrives with a body (as an upstream/proxy might send).
+// See https://github.com/Azure/azure-functions-nodejs-library/issues/458
+function sendRequestWithBody(
+    url: string,
+    method: 'GET' | 'HEAD',
+    body: string
+): Promise<{ status: number; body: string }> {
+    return new Promise((resolve, reject) => {
+        const parsed = new URL(url);
+        const req = http.request(
+            {
+                hostname: parsed.hostname,
+                port: parsed.port,
+                path: `${parsed.pathname}${parsed.search}`,
+                method,
+                headers: {
+                    'content-type': 'text/plain',
+                    'content-length': Buffer.byteLength(body),
+                },
+            },
+            (res) => {
+                let data = '';
+                res.setEncoding('utf8');
+                res.on('data', (chunk) => {
+                    data += chunk;
+                });
+                res.on('end', () => resolve({ status: res.statusCode ?? 0, body: data }));
+            }
+        );
+        req.on('error', reject);
+        req.write(body);
+        req.end();
+    });
+}
+
 describe('http', () => {
     afterEach(() => {
         funcCliSettings.hideOutput = false;
@@ -124,6 +161,27 @@ describe('http', () => {
         }
     });
 
+    // Regression test for https://github.com/Azure/azure-functions-nodejs-library/issues/458
+    // A GET/HEAD request that arrives with a body must reach the handler instead of crashing the worker
+    // while constructing the WHATWG Request (which rejects bodies for GET/HEAD). The name is passed via
+    // query so the response is deterministic across models and streaming modes; a status of 200 (rather
+    // than a 500) confirms the request reached the handler.
+    describe('GET/HEAD request with a body (issue #458)', () => {
+        it('GET with a body reaches the handler', async () => {
+            const url = getFuncUrl('helloWorld', { name: 'getWithBody' });
+            const { status, body } = await sendRequestWithBody(url, 'GET', 'this GET body must not crash the worker');
+            expect(status).to.equal(200);
+            expect(body).to.equal('Hello, getWithBody!');
+        });
+
+        it('HEAD with a body reaches the handler', async () => {
+            const url = getFuncUrl('helloWorld', { name: 'headWithBody' });
+            // HEAD responses carry no body, so we only assert the request reached the handler (status 200).
+            const { status } = await sendRequestWithBody(url, 'HEAD', 'this HEAD body must not crash the worker');
+            expect(status).to.equal(200);
+        });
+    });
+
     // Use a connection pool to avoid flaky test failures due to various connection limits (Mac in particular seems to have a low limit)
     // NOTE: The node-fetch package has a bug starting in 2.6.8 related to keep alive agents, so we have to use 2.6.7
     // https://github.com/node-fetch/node-fetch/issues/1767