CC: Remove auto-retry on code 429

13db803bc666b1eedfc745741c222e0de4051d78

Cohee <18619528+Cohee1207@users.noreply.github.com>

1 files changed, +4 -13Showing whitespace changes
src/endpoints/backends/chat-completions.js+4 -13
@@ -1792,7 +1792,7 @@ router.post('/generate', function (request, response) {
17921792 signal: controller.signal,
17931793 };
17941794
17951795 console.debug('Chat Completion request:', requestBody);
17961796
17971797 makeRequest(config, response, request);
17981798
@@ -1801,10 +1801,8 @@ router.post('/generate', function (request, response) {
18011801 * @param {import('node-fetch').RequestInit} config Fetch config
18021802 * @param {express.Response} response Express response
18031803 * @param {express.Request} request Express request
1804- * @param {Number} retries Number of retries left
1805- * @param {Number} timeout Request timeout in ms
18061804 */
18071805 async function makeRequest(config, response, request, retries = 5, timeout = 5000) {
18081806 try {
18091807 controller.signal.throwIfAborted();
18101808 const fetchResponse = await fetch(endpointUrl, config);
@@ -1819,14 +1817,7 @@ router.post('/generate', function (request, response) {
18191817 /** @type {any} */
18201818 let json = await fetchResponse.json();
18211819 response.send(json);
18221820 console.debug('Chat Completion response:', json);
1823- console.debug(json?.choices?.[0]?.message);
1824- } else if (fetchResponse.status === 429 && retries > 0) {
1825- console.warn(`Out of quota, retrying in ${Math.round(timeout / 1000)}s`);
1826- setTimeout(() => {
1827- timeout *= 2;
1828- makeRequest(config, response, request, retries - 1, timeout);
1829- }, timeout);
18301821 } else {
18311822 await handleErrorResponse(fetchResponse);
18321823 }
@@ -1858,7 +1849,7 @@ router.post('/generate', function (request, response) {
18581849 if (!response.headersSent) {
18591850 response.send({ error: { message }, quota_error: quota_error });
18601851 } else if (!response.writableEnded) {
18611852 response.write(errorResponseresponseText);
18621853 } else {
18631854 response.end();
18641855 }