fix(sse): preserve original body for semantic cache signature — fixes 0% hit rate (#9775)

The semantic cache signature (generateSignature) was computed over different
bodies at read-time vs write-time in handleChatCore. The cache read at Phase
9.1 uses the original body, but the writes at Phase 9.1 (non-streaming) and
Phase 9.2 (streaming) used the body after sanitizeChatRequestBody() and
injectMemoryAndSkills() mutated messages. Since the digest includes messages,
every request stored under a key no later request would look up — 0% hit
rate, every request billed.

Fix: snapshot bodyForCacheWrite right after the cache read and use it for
both write paths, so the write-time signature equals the read-time one.

TDD: tests/unit/cache-signature-roundtrip.test.ts proves the mutated body
produces a different signature (bug) and the preserved snapshot produces an
identical one (fix).

Co-authored-by: diegosouzapw <diegosouzapw@users.noreply.github.com>
This commit is contained in:
Diego Rodrigues de Sa e Souza
2026-08-10 20:10:16 -03:00
committed by GitHub
parent 5e02cf7623
commit 0481246bf4
2 changed files with 109 additions and 2 deletions

View File

@@ -1037,6 +1037,13 @@ export async function handleChatCore({
log?.debug?.("FORMAT", `${sourceFormat}${targetFormat} | stream=${stream}`);
// Preserve original body for cache signature — the body variable is mutated
// multiple times below (sanitization, memory/skills injection) before the
// cache store path runs at Phase 9.1 (non-streaming) / Phase 9.2 (streaming).
// Without this snapshot, the write-time signature differs from the read-time
// one, producing 0% hit rate. (#cache-signature-asymmetry)
const bodyForCacheWrite = body;
// ── Phase 9.1: Semantic cache check (temp=0, any streaming mode) ──
const cacheHit = await checkSemanticCache({
semanticCacheEnabled,
@@ -4568,7 +4575,7 @@ export async function handleChatCore({
// ── Phase 9.1: Cache store (non-streaming, temp=0) ──
storeSemanticCacheResponse({
enabled: semanticCacheEnabled,
body,
body: bodyForCacheWrite,
headers: clientRawRequest?.headers,
translatedResponse,
model,
@@ -4922,7 +4929,7 @@ export async function handleChatCore({
enabled: semanticCacheEnabled,
streamStatus,
streamResponseBody,
body,
body: bodyForCacheWrite,
headers: clientRawRequest?.headers,
model,
apiKeyId: apiKeyInfo?.id ?? undefined,

View File

@@ -0,0 +1,100 @@
// TDD regression guard: semantic cache read-time vs write-time signature must match.
// Without bodyForCacheWrite, the body variable is mutated between the cache read
// (Phase 9.1, chatCore.ts:1055) and the cache write (Phase 9.1 non-streaming:4540 /
// Phase 9.2 streaming:4896) by sanitizeChatRequestBody() and injectMemoryAndSkills().
// The digest includes messages, so mutations produce a different key → 0% hit rate.
import test from "node:test";
import assert from "node:assert/strict";
const { generateSignature } = await import("../../src/lib/semanticCache.ts");
const BASE_BODY = {
model: "gpt-4o",
messages: [{ role: "user", content: "What is the capital of France?" }],
temperature: 0,
top_p: 1,
};
// Mutations that handleChatCore applies between cache read and cache write.
function simulateSanitizeChatRequestBody(body: Record<string, unknown>): Record<string, unknown> {
return { ...body, _sanitized: true };
}
function simulateInjectMemoryAndSkills(body: Record<string, unknown>): Record<string, unknown> {
return {
...body,
messages: [
{ role: "system", content: "[Memory: user likes programming]" },
...(body.messages as Array<unknown>),
],
};
}
test("cache signature must be identical before and after body mutations", async () => {
// Compute signature at read-time (original body, line 1055).
const readSignature = await generateSignature(
BASE_BODY.model,
BASE_BODY.messages,
BASE_BODY.temperature,
BASE_BODY.top_p
);
// Simulate the mutations that happen between cache read and cache write.
let mutatedBody = simulateSanitizeChatRequestBody(BASE_BODY);
mutatedBody = simulateInjectMemoryAndSkills(mutatedBody);
// Compute signature at write-time using the mutated body (the bug).
const writeSignatureMutated = await generateSignature(
mutatedBody.model,
mutatedBody.messages,
mutatedBody.temperature,
mutatedBody.top_p
);
// Compute signature at write-time using the preserved snapshot (the fix).
const writeSignaturePreserved = await generateSignature(
BASE_BODY.model,
BASE_BODY.messages,
BASE_BODY.temperature,
BASE_BODY.top_p
);
// The mutated signature MUST differ — this is the bug.
assert.notStrictEqual(
readSignature,
writeSignatureMutated,
"BUG: mutated body produces different signature — cache has 0% hit rate"
);
// The preserved snapshot MUST match — this is the fix.
assert.strictEqual(
readSignature,
writeSignaturePreserved,
"FIX: preserved snapshot must produce the same signature as read-time"
);
});
test("cache signature with memory injection changes messages", async () => {
const bodyWithMemory = {
...BASE_BODY,
messages: [
{ role: "system", content: "[Memory: context from previous conversation]" },
...BASE_BODY.messages,
],
};
const sigOriginal = await generateSignature(
BASE_BODY.model,
BASE_BODY.messages,
BASE_BODY.temperature,
BASE_BODY.top_p
);
const sigWithMemory = await generateSignature(
bodyWithMemory.model,
bodyWithMemory.messages,
bodyWithMemory.temperature,
bodyWithMemory.top_p
);
assert.notStrictEqual(sigOriginal, sigWithMemory);
});