Skip to content

fix(ai-gateway): close terminal SSE streams #4882

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
chrarnoldus merged 2 commits into from
Jul 30, 2026
Merged

fix(ai-gateway): close terminal SSE streams #4882

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
c85923a
fix(ai-gateway): close terminal Responses streams
chrarnoldus Jul 30, 2026
e6b4826
fix(ai-gateway): close all terminal SSE streams
chrarnoldus Jul 30, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
91 changes: 89 additions & 2 deletions apps/web/src/lib/rewriteModelResponse.test.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -39,6 +39,22 @@ function sseResponse(body: string, status = 200): Response {
});
}

function hangingSseResponse(body: string): { response: Response; cancel: jest.Mock } {
const encoder = new TextEncoder();
const cancel = jest.fn();
const stream = new ReadableStream<Uint8Array>({
start(controller) {
controller.enqueue(encoder.encode(body));
},
cancel,
});

return {
response: new Response(stream, { headers: { 'content-type': 'text/event-stream' } }),
cancel,
};
}

function failingResponse(contentType: string, errorName: string, initialBody?: string): Response {
const encoder = new TextEncoder();
let pullCount = 0;
Expand Down Expand Up @@ -302,6 +318,24 @@ describe('rewriteModelResponse_ChatCompletions', () => {
expect(dataPayloads(sse)).toContain('[DONE]');
});

test('cancels upstream and closes immediately after [DONE]', async () => {
const capture = makeCapture();
const body =
'data: {"id":"gen-chat","model":"upstream-model","choices":[]}\n\n' +
'data: [DONE]\n\n' +
'data: {"id":"ignored","model":"upstream-model","choices":[]}\n\n';
const { response: upstream, cancel } = hangingSseResponse(body);

const result = await rewriteModelResponse_ChatCompletions(upstream, true, capture, null);
const sse = await readOutputStream(result);

expect(dataObjects(sse)).toEqual([{ id: 'gen-chat', model: 'upstream-model', choices: [] }]);
expect(dataPayloads(sse)).toContain('[DONE]');
expect(cancel).toHaveBeenCalledTimes(1);
expect(capture.setBody).toHaveBeenCalledWith(body);
expect(capture.setReadError).not.toHaveBeenCalled();
});

test('adds an empty choices array and strips cost on usage-only chunks', async () => {
const upstream = sseResponse(
'data: {"model":"upstream-model","usage":{"cost":1,"is_byok":true,"prompt_tokens":4,"completion_tokens":2,"total_tokens":6,"prompt_tokens_details":{}}}\n\n'
Expand Down Expand Up @@ -464,6 +498,40 @@ describe('rewriteModelResponse_Messages', () => {

expect(dataPayloads(sse)).not.toContain('[DONE]');
});

test('cancels upstream and closes immediately after message_stop', async () => {
const capture = makeCapture();
const body =
'event: message_stop\ndata: {"type":"message_stop"}\n\n' +
'event: message_delta\ndata: {"type":"message_delta","usage":{"output_tokens":10},"delta":{}}\n\n';
const { response: upstream, cancel } = hangingSseResponse(body);

const result = await rewriteModelResponse_Messages(upstream, true, capture, null);
const sse = await readOutputStream(result);

expect(dataObjects(sse)).toEqual([{ type: 'message_stop' }]);
expect(dataPayloads(sse)).not.toContain('[DONE]');
expect(cancel).toHaveBeenCalledTimes(1);
expect(capture.setBody).toHaveBeenCalledWith(body);
expect(capture.setReadError).not.toHaveBeenCalled();
});

test('cancels upstream and closes immediately after a compatible [DONE] sentinel', async () => {
const capture = makeCapture();
const body =
'data: [DONE]\n\n' +
'event: message_delta\ndata: {"type":"message_delta","usage":{"output_tokens":10},"delta":{}}\n\n';
const { response: upstream, cancel } = hangingSseResponse(body);

const result = await rewriteModelResponse_Messages(upstream, true, capture, null);
const sse = await readOutputStream(result);

expect(dataObjects(sse)).toEqual([]);
expect(dataPayloads(sse)).toEqual(['[DONE]']);
expect(cancel).toHaveBeenCalledTimes(1);
expect(capture.setBody).toHaveBeenCalledWith(body);
expect(capture.setReadError).not.toHaveBeenCalled();
});
});

describe('rewriteModelResponse_Responses', () => {
Expand Down Expand Up @@ -526,7 +594,7 @@ describe('rewriteModelResponse_Responses', () => {
expect(json.usage.prompt_tokens_details.cached_tokens).toBe(0);
});

test('strips the nested response usage in stream events and emits [DONE]', async () => {
test('strips nested response usage and closes after the completed event', async () => {
const upstream = sseResponse(
'event: response.completed\n' +
'data: {"type":"response.completed","response":{"model":"upstream-model","usage":{"cost":0.5,"is_byok":true,"prompt_tokens":3,"completion_tokens":1,"total_tokens":4,"prompt_tokens_details":{"cached_tokens":1}}}}\n\n' +
Expand All @@ -552,8 +620,27 @@ describe('rewriteModelResponse_Responses', () => {
expect(event.response.usage.is_byok).toBeUndefined();
expect(event.response.usage.prompt_tokens_details.cached_tokens).toBe(1);
expect(sse).toContain('event: response.completed');
expect(dataPayloads(sse)).toContain('[DONE]');
expect(dataPayloads(sse)).not.toContain('[DONE]');
});

test.each(['response.completed', 'response.incomplete', 'response.failed'])(
'forwards %s, cancels upstream, and closes without waiting for EOF',
async type => {
const capture = makeCapture();
const body =
`event: ${type}\ndata: ${JSON.stringify({ type })}\n\n` +
'event: response.output_text.delta\ndata: {"type":"response.output_text.delta","delta":"ignored"}\n\n';
const { response: upstream, cancel } = hangingSseResponse(body);

const result = await rewriteModelResponse_Responses(upstream, true, capture, null);
const sse = await readOutputStream(result);

expect(dataObjects(sse)).toEqual([{ type }]);
expect(cancel).toHaveBeenCalledTimes(1);
expect(capture.setBody).toHaveBeenCalledWith(body);
expect(capture.setReadError).not.toHaveBeenCalled();
}
);
});

function makeLogging(overrides?: Partial<RequestLoggingParams>): RequestLoggingParams {
Expand Down
57 changes: 53 additions & 4 deletions apps/web/src/lib/rewriteModelResponse.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -254,6 +254,7 @@ async function rewriteSseStream(
parser: ReturnType<typeof createParser>,
controller: ReadableStreamDefaultController<string>,
doneReceived: () => boolean,
terminalEventReceived: () => boolean,
serializeError: (error: ResponseReadError) => string,
onFinally: () => void,
vercelRequestId: string | null | undefined,
Expand All @@ -268,6 +269,12 @@ async function rewriteSseStream(
error,
capturedChunks && capturedChunks.length > 0 ? capturedChunks.join('') : undefined
);
const settleBody = () => {
if (capturedChunks) {
capturedChunks.push(decoder.decode());
capture?.setBody(capturedChunks.join(''));
}
};
try {
while (true) {
const { done, value } = await reader.read();
Expand All @@ -279,15 +286,29 @@ async function rewriteSseStream(
controller.enqueue('data: [DONE]\n\n');
}
controller.close();
if (capturedChunks) {
capturedChunks.push(decoder.decode());
capture?.setBody(capturedChunks.join(''));
}
settleBody();
return;
}
const chunk = decoder.decode(value, { stream: true });
capturedChunks?.push(chunk);
parser.feed(chunk);
if (terminalEventReceived()) {
if (doneReceived()) {
controller.enqueue('data: [DONE]\n\n');
}
const cancellation = reader.cancel();
controller.close();
settleBody();
try {
await cancellation;
} catch (error) {
errorExceptInTest(
'[rewriteModelResponse] failed to cancel terminal upstream stream',
error
);
}
return;
}
}
} catch (error) {
const responseReadError = getResponseReadError(error, vercelRequestId);
Expand Down Expand Up @@ -377,6 +398,9 @@ export async function rewriteModelResponse_ChatCompletions(
const progress = createStreamProgressLogger();
const parser = createParser({
onEvent(event: EventSourceMessage) {
if (doneReceived) {
return;
}
progress.eventProcessed();
if (event.data === '[DONE]') {
doneReceived = true;
Expand Down Expand Up @@ -412,6 +436,9 @@ export async function rewriteModelResponse_ChatCompletions(
controller.enqueue(eventLine + 'data: ' + JSON.stringify(json) + '\n\n');
},
onComment() {
if (doneReceived) {
return;
}
controller.enqueue(': KILO PROCESSING\n\n');
},
});
Expand All @@ -421,6 +448,7 @@ export async function rewriteModelResponse_ChatCompletions(
parser,
controller,
() => doneReceived,
() => doneReceived,
responseReadError =>
'data: ' +
JSON.stringify({
Expand Down Expand Up @@ -526,10 +554,14 @@ export async function rewriteModelResponse_Messages(
}

let doneReceived = false;
let terminalEventReceived = false;
let generationId: string | undefined;
const progress = createStreamProgressLogger();
const parser = createParser({
onEvent(event: EventSourceMessage) {
if (doneReceived || terminalEventReceived) {
return;
}
progress.eventProcessed();
if (event.data === '[DONE]') {
doneReceived = true;
Expand Down Expand Up @@ -563,8 +595,12 @@ export async function rewriteModelResponse_Messages(

const eventLine = event.event ? 'event: ' + event.event + '\n' : '';
controller.enqueue(eventLine + 'data: ' + JSON.stringify(json) + '\n\n');
terminalEventReceived = json.type === 'message_stop';
},
onComment() {
if (doneReceived || terminalEventReceived) {
return;
}
controller.enqueue(': KILO PROCESSING\n\n');
},
});
Expand All @@ -574,6 +610,7 @@ export async function rewriteModelResponse_Messages(
parser,
controller,
() => doneReceived,
() => doneReceived || terminalEventReceived,
responseReadError =>
'event: error\n' +
'data: ' +
Expand Down Expand Up @@ -662,11 +699,15 @@ export async function rewriteModelResponse_Responses(
}

let doneReceived = false;
let terminalEventReceived = false;
let generationId: string | undefined;
let nextSequenceNumber = 0;
const progress = createStreamProgressLogger();
const parser = createParser({
onEvent(event: EventSourceMessage) {
if (doneReceived || terminalEventReceived) {
return;
}
progress.eventProcessed();
if (event.data === '[DONE]') {
doneReceived = true;
Expand All @@ -690,8 +731,15 @@ export async function rewriteModelResponse_Responses(
}
const eventLine = event.event ? 'event: ' + event.event + '\n' : '';
controller.enqueue(eventLine + 'data: ' + JSON.stringify(json) + '\n\n');
terminalEventReceived =
json.type === 'response.completed' ||
json.type === 'response.incomplete' ||
json.type === 'response.failed';
},
onComment() {
if (doneReceived || terminalEventReceived) {
return;
}
controller.enqueue(': KILO PROCESSING\n\n');
},
});
Expand All @@ -701,6 +749,7 @@ export async function rewriteModelResponse_Responses(
parser,
controller,
() => doneReceived,
() => doneReceived || terminalEventReceived,
responseReadError =>
'event: error\n' +
'data: ' +
Expand Down