retry stream capacity errors in gateway

This commit is contained in:
2026-06-29 20:54:25 +08:00
parent ab0572baf6
commit 1d7016bf65
3 changed files with 371 additions and 117 deletions
+45
View File
@@ -80,6 +80,21 @@ function createTerminatedSseResponse(res, chunks, destroyDelayMs = 20) {
}, destroyDelayMs);
}
function createCapacityErrorSseResponse(
res,
message = "Selected model is at capacity. Please try a different model.",
intervalMs = 20,
) {
createSseResponse(
res,
[
'event: error\n',
`data: ${JSON.stringify({ error: { message, type: "server_error" } })}\n\n`,
],
intervalMs,
);
}
function startFakeUpstream(port) {
const failBeforeResponseCounts = new Map();
const capacityBeforeSuccessCounts = new Map();
@@ -143,6 +158,13 @@ function startFakeUpstream(port) {
const capacityCount = (capacityBeforeSuccessCounts.get(capacityKey) || 0) + 1;
capacityBeforeSuccessCounts.set(capacityKey, capacityCount);
if (capacityCount <= parsed.test_capacity_before_success_times) {
if (parsed.stream) {
createCapacityErrorSseResponse(
res,
parsed.test_capacity_message || "Selected model is at capacity. Please try a different model.",
);
return;
}
createJsonResponse(
res,
parsed.test_capacity_status ?? 503,
@@ -157,6 +179,13 @@ function startFakeUpstream(port) {
return;
}
}
if (parsed.stream && parsed.test_capacity_error) {
createCapacityErrorSseResponse(
res,
parsed.test_capacity_message || "Selected model is at capacity. Please try a different model.",
);
return;
}
if (parsed.stream) {
createSseResponse(res, [
'data: {"type":"response.output_text.delta","delta":"hello"}\n\n',
@@ -434,6 +463,22 @@ async function run() {
"stream+capacity error 返回体未标记 retry trigger",
);
const streamCapacityRecoveredResponse = await fetch(`http://127.0.0.1:${gatewayPort}/responses`, {
method: "POST",
headers: { "content-type": "application/json" },
body: JSON.stringify({ stream: true, test_capacity_before_success_times: 2, test_reasoning_tokens: 128 }),
});
const streamCapacityRecoveredText = await streamCapacityRecoveredResponse.text();
assert(streamCapacityRecoveredResponse.status === 200, `stream capacity 抖动后未自动恢复: ${streamCapacityRecoveredResponse.status}`);
assert(streamCapacityRecoveredText.includes("hello"), "stream capacity 恢复后未拿到正常 SSE 内容");
const requestsAfterStreamCapacityRecoveryResponse = await fetch(`http://127.0.0.1:${gatewayPort}/__codex_retry_gateway/api/requests?limit=20`);
const requestsAfterStreamCapacityRecovery = await requestsAfterStreamCapacityRecoveryResponse.json();
const streamCapacityRecoveredEntry = requestsAfterStreamCapacityRecovery?.entries?.find(
(entry) => entry.path === "/responses" && entry.status_code === 200 && entry.response_stream && entry.upstream_attempt_count >= 3,
);
assert(streamCapacityRecoveredEntry, "stream capacity 抖动恢复后的请求记录未保留重试次数");
for (const streamPath of [
"/responses",
"/v1/responses",