support streaming and closes properly on site targets

This commit is contained in:
Owen
2026-08-06 17:15:11 -04:00
parent 6c28c5f383
commit 22f2990f56
4 changed files with 92 additions and 18 deletions
+4 -3
View File
@@ -62,9 +62,10 @@ export function joinUpstreamUrl(baseUrl: string, path: string): string {
} }
function pathFromRequest(req: Request): string { function pathFromRequest(req: Request): string {
// Prefer originalUrl path (includes mounted path) over req.path when available. // Prefer originalUrl (includes mounted path) over req.url when available.
const raw = // Query string is preserved - some providers use it to select the
req.originalUrl?.split("?")[0] || req.url?.split("?")[0] || req.path; // streaming response format (e.g. Gemini's `?alt=sse`).
const raw = req.originalUrl || req.url || req.path;
return raw.startsWith("/") ? raw : `/${raw}`; return raw.startsWith("/") ? raw : `/${raw}`;
} }
+14
View File
@@ -7,6 +7,7 @@ type UpstreamFetchInit = {
headers: Record<string, string>; headers: Record<string, string>;
body?: string; body?: string;
skipTlsVerification?: boolean; skipTlsVerification?: boolean;
signal?: AbortSignal;
}; };
const insecureHttpsAgent = new https.Agent({ const insecureHttpsAgent = new https.Agent({
@@ -25,6 +26,11 @@ export function aiGatewayUpstreamFetch(
isHttps && init.skipTlsVerification ? insecureHttpsAgent : undefined; isHttps && init.skipTlsVerification ? insecureHttpsAgent : undefined;
return new Promise((resolve, reject) => { return new Promise((resolve, reject) => {
if (init.signal?.aborted) {
reject(init.signal.reason ?? new Error("Request aborted"));
return;
}
const req = lib.request( const req = lib.request(
url, url,
{ {
@@ -60,6 +66,14 @@ export function aiGatewayUpstreamFetch(
req.on("error", reject); req.on("error", reject);
if (init.signal) {
const onAbort = () => req.destroy(init.signal!.reason);
init.signal.addEventListener("abort", onAbort, { once: true });
req.on("close", () =>
init.signal!.removeEventListener("abort", onAbort)
);
}
if (init.body !== undefined) { if (init.body !== undefined) {
req.write(init.body); req.write(init.body);
} }
+33 -6
View File
@@ -592,15 +592,33 @@ export async function handleAiGatewayProxy(
skipTlsVerification: provider.skipTlsVerification skipTlsVerification: provider.skipTlsVerification
}); });
// Cancel the upstream request (and, transitively, anything it fans
// out to) if the client goes away before we're done - otherwise a
// client-cancelled streaming chat completion keeps running upstream
// to completion, wasting the connection and any per-token billing.
const abortController = new AbortController();
const onClientClose = () => {
if (!res.writableEnded) {
abortController.abort();
}
};
res.on("close", onClientClose);
let upstreamRes: globalThis.Response; let upstreamRes: globalThis.Response;
try { try {
upstreamRes = await aiGatewayUpstreamFetch(targetUrl, { upstreamRes = await aiGatewayUpstreamFetch(targetUrl, {
method: "POST", method: "POST",
headers, headers,
body, body,
skipTlsVerification: provider.skipTlsVerification skipTlsVerification: provider.skipTlsVerification,
signal: abortController.signal
}); });
} catch (fetchError) { } catch (fetchError) {
res.off("close", onClientClose);
if (abortController.signal.aborted) {
// Client already disconnected; nothing left to respond to.
return;
}
logger.error({ logger.error({
message: "AI gateway upstream fetch failed", message: "AI gateway upstream fetch failed",
url: targetUrl, url: targetUrl,
@@ -628,14 +646,23 @@ export async function handleAiGatewayProxy(
if (isStream && upstreamRes.body) { if (isStream && upstreamRes.body) {
res.flushHeaders(); res.flushHeaders();
const reader = upstreamRes.body.getReader(); const reader = upstreamRes.body.getReader();
while (true) { try {
const { done, value } = await reader.read(); while (!abortController.signal.aborted) {
if (done) break; const { done, value } = await reader.read();
res.write(value); if (done) break;
res.write(value);
}
} finally {
await reader.cancel().catch(() => {});
res.off("close", onClientClose);
} }
return res.end(); if (!res.writableEnded) {
res.end();
}
return;
} }
res.off("close", onClientClose);
const text = await upstreamRes.text(); const text = await upstreamRes.text();
return res.send(text); return res.send(text);
} catch (error) { } catch (error) {
+41 -9
View File
@@ -128,8 +128,10 @@ function pickTarget(
} }
function pathFromRequest(req: Request): string { function pathFromRequest(req: Request): string {
const raw = // Query string is preserved - some providers use it to select the
req.originalUrl?.split("?")[0] || req.url?.split("?")[0] || req.path; // streaming response format (e.g. Gemini's `?alt=sse`), and gerbil's
// /router/* forwards it through untouched.
const raw = req.originalUrl || req.url || req.path;
return raw.startsWith("/") ? raw : `/${raw}`; return raw.startsWith("/") ? raw : `/${raw}`;
} }
@@ -205,14 +207,32 @@ export async function proxyAiGatewayToSiteTarget(
body: req.body body: req.body
}); });
// Cancel the request to gerbil (which cascades to gerbil cancelling its
// proxied request to the actual site target, since gerbil's reverse
// proxy derives the outbound request's context from the inbound one) if
// the client goes away before we're done.
const abortController = new AbortController();
const onClientClose = () => {
if (!res.writableEnded) {
abortController.abort();
}
};
res.on("close", onClientClose);
let upstreamRes: globalThis.Response; let upstreamRes: globalThis.Response;
try { try {
upstreamRes = await fetch(gerbilUrl, { upstreamRes = await fetch(gerbilUrl, {
method: "POST", method: "POST",
headers, headers,
body body,
signal: abortController.signal
}); });
} catch (fetchError) { } catch (fetchError) {
res.off("close", onClientClose);
if (abortController.signal.aborted) {
// Client already disconnected; nothing left to respond to.
return;
}
logger.error({ logger.error({
message: "AI gateway target proxy request failed", message: "AI gateway target proxy request failed",
url: gerbilUrl, url: gerbilUrl,
@@ -232,7 +252,11 @@ export async function proxyAiGatewayToSiteTarget(
const contentType = upstreamRes.headers.get("content-type") || ""; const contentType = upstreamRes.headers.get("content-type") || "";
const isStream = const isStream =
req.body?.stream === true || req.body?.stream === true ||
contentType.includes("text/event-stream"); contentType.includes("text/event-stream") ||
pathFromRequest(req).includes("streamGenerateContent") ||
pathFromRequest(req).includes("streamRawPredict") ||
pathFromRequest(req).includes("converse-stream") ||
pathFromRequest(req).includes("invoke-with-response-stream");
res.status(upstreamRes.status); res.status(upstreamRes.status);
res.setHeader("Content-Type", contentType || "application/json"); res.setHeader("Content-Type", contentType || "application/json");
@@ -240,15 +264,23 @@ export async function proxyAiGatewayToSiteTarget(
if (isStream && upstreamRes.body) { if (isStream && upstreamRes.body) {
res.flushHeaders(); res.flushHeaders();
const reader = upstreamRes.body.getReader(); const reader = upstreamRes.body.getReader();
while (true) { try {
const { done, value } = await reader.read(); while (!abortController.signal.aborted) {
if (done) break; const { done, value } = await reader.read();
res.write(value); if (done) break;
res.write(value);
}
} finally {
await reader.cancel().catch(() => {});
res.off("close", onClientClose);
}
if (!res.writableEnded) {
res.end();
} }
res.end();
return; return;
} }
res.off("close", onClientClose);
const text = await upstreamRes.text(); const text = await upstreamRes.text();
res.send(text); res.send(text);
} }