From ee3546a6cedd044cbbb45a3937868f8b7a416d08 Mon Sep 17 00:00:00 2001
From: Ravi Tharuma <25951435+RaviTharuma@users.noreply.github.com>
Date: Mon, 20 Jul 2026 20:56:02 +0200
Subject: [PATCH] perf: reduce long-context request copies (#7862)

Co-authored-by: Ravi Tharuma <RaviTharuma@users.noreply.github.com>
---
 src/app/api/v1/chat/completions/route.ts      |   5 +-
 .../chat-completions-parse-once-7847.test.ts  | 112 ++++++++++++++++++
 2 files changed, 114 insertions(+), 3 deletions(-)
 create mode 100644 tests/unit/chat-completions-parse-once-7847.test.ts

diff --git a/src/app/api/v1/chat/completions/route.ts b/src/app/api/v1/chat/completions/route.ts
index 2915d4730b..d2ce928a65 100644
--- a/src/app/api/v1/chat/completions/route.ts
+++ b/src/app/api/v1/chat/completions/route.ts
@@ -65,7 +65,7 @@ export async function POST(request) {
   }
 
   // Heap-pressure-aware admission: shed a large body with 503 (or 413 if pathological)
-  // BEFORE the request is cloned + JSON-parsed below. A large coding-agent compact body
+  // BEFORE the request is JSON-parsed below. A large coding-agent compact body
   // amplifies into hundreds of MB of transient JS objects on the combo path; under a
   // burst of concurrent compacts that stacks past the V8 heap ceiling and OOM-crashes the
   // whole process. Shedding the marginal request here turns a pod-wide crash into a single
@@ -90,8 +90,7 @@ export async function POST(request) {
   // residency on the hot path and fed the OOM crash-loop (#4380).
   let parsedBody = null;
   try {
-    const cloned = request.clone();
-    parsedBody = await cloned.json().catch(() => null);
+    parsedBody = await request.json().catch(() => null);
     if (parsedBody) {
       const { blocked, result } = injectionGuard(parsedBody);
       if (blocked) {
diff --git a/tests/unit/chat-completions-parse-once-7847.test.ts b/tests/unit/chat-completions-parse-once-7847.test.ts
new file mode 100644
index 0000000000..9b61a3a025
--- /dev/null
+++ b/tests/unit/chat-completions-parse-once-7847.test.ts
@@ -0,0 +1,112 @@
+import test from "node:test";
+import assert from "node:assert/strict";
+import fs from "node:fs";
+import os from "node:os";
+import path from "node:path";
+
+// #7847: /v1/chat/completions already threads its parsed body into handleChat, so cloning
+// the Request before parsing tees and retains an unused serialized-body branch. These tests
+// pin the memory-sensitive contract: consume the original body once without cloning it.
+
+const TEST_DATA_DIR = fs.mkdtempSync(path.join(os.tmpdir(), "omniroute-chat-parse-once-7847-"));
+process.env.DATA_DIR = TEST_DATA_DIR;
+
+const core = await import("../../src/lib/db/core.ts");
+const chatRoute = await import("../../src/app/api/v1/chat/completions/route.ts");
+const { resolveChatRequestBody } = await import("../../src/sse/handlers/requestBody.ts");
+
+function makeCountingRequest(body: string) {
+  const request = new Request("http://localhost/v1/chat/completions", {
+    method: "POST",
+    headers: { "Content-Type": "application/json" },
+    body,
+  });
+
+  let jsonCalls = 0;
+  let cloneCalls = 0;
+  const originalJson = request.json.bind(request);
+  const originalClone = request.clone.bind(request);
+
+  Object.defineProperties(request, {
+    json: {
+      value: async () => {
+        jsonCalls++;
+        return originalJson();
+      },
+    },
+    clone: {
+      value: () => {
+        cloneCalls++;
+        return originalClone();
+      },
+    },
+  });
+
+  return {
+    request,
+    jsonCalls: () => jsonCalls,
+    cloneCalls: () => cloneCalls,
+  };
+}
+
+function validBody(stream: boolean) {
+  return JSON.stringify({
+    model: "openai/gpt-4.1",
+    messages: [{ role: "user", content: "Reply with OK only." }],
+    stream,
+  });
+}
+
+test("#7847 non-streaming requests parse the original body once without cloning", async () => {
+  const counting = makeCountingRequest(validBody(false));
+
+  await chatRoute.POST(counting.request);
+
+  assert.equal(counting.jsonCalls(), 1, "the original request must be parsed exactly once");
+  assert.equal(counting.cloneCalls(), 0, "the request body stream must not be teed");
+});
+
+test("#7847 streaming requests parse the original body once without cloning", async () => {
+  const counting = makeCountingRequest(validBody(true));
+
+  const response = await chatRoute.POST(counting.request);
+  await response.text();
+
+  assert.equal(counting.jsonCalls(), 1, "early keepalive must reuse the parsed body");
+  assert.equal(counting.cloneCalls(), 0, "streaming must not retain an unread body branch");
+});
+
+test("#7847 malformed JSON still returns 400 after the original body is consumed", async () => {
+  const counting = makeCountingRequest('{"model":');
+
+  const response = await chatRoute.POST(counting.request);
+
+  assert.equal(response.status, 400);
+  assert.equal(counting.cloneCalls(), 0, "invalid JSON must not require cloning the body stream");
+});
+
+test("#7847 downstream body resolution preserves the parsed object's identity", async () => {
+  const parsedBody = {
+    model: "openai/gpt-4.1",
+    messages: [{ role: "user", content: "identity sentinel" }],
+  };
+  let downstreamJsonCalls = 0;
+
+  const resolved = await resolveChatRequestBody(
+    {
+      json: async () => {
+        downstreamJsonCalls++;
+        return { unexpected: true };
+      },
+    },
+    parsedBody
+  );
+
+  assert.equal(resolved, parsedBody, "downstream must reuse the exact parsed object");
+  assert.equal(downstreamJsonCalls, 0, "downstream must not parse or materialize another body");
+});
+
+test.after(() => {
+  core.resetDbInstance();
+  fs.rmSync(TEST_DATA_DIR, { recursive: true, force: true });
+});