Handling Errors in Streaming Responses
Handling Errors in Streaming Responses
Jul 14, 2024
·
1 min read
·
148
Words
·
-Views
-Comments
AI APIs often return server-sent streams. When the server reports an error, the payload still arrives as a stream, so error handling differs from typical JSON responses.
Example
Using Axios as an example:
axios.post('http://api.com/v1/chat/completions', data, {
headers: {
Authorization: `Bearer xxxx`,
'Content-Type': 'application/json',
},
responseType: 'stream',
})
.then(res => res.data)
.catch(async (e) => {
let error;
if (e.isAxiosError) {
let streamString = '';
await new Promise(resolve => {
e.response.data
.on('data', (utf8Chunk) => {
streamString += utf8Chunk;
})
.on('end', () => {
error = JSON.parse(streamString);
resolve();
});
});
} else {
error = e;
}
throw error;
});
In streaming mode you must consume the error payload through the same evented interface as normal data. Wrapping the listener with a Promise ensures you capture the full string before parsing and rethrowing.
Final Thoughts
When dealing with streaming APIs, remember that error paths are still streams—structure your error handling accordingly.
