503 Service Temporarily Unavailable for some users

API
1

Hi folks,

I’m using the API with gpt-3.5-turbo and as of this morning I started getting a lot of timeouts “cf_*” when calling the API.

This afternoon started to return “empty” responses with 503 status.

I’ve investigated and noticed that if I log in to my web app with user id “1”, I get the 503. However, when I log in with user id “2”, the issue seems to be gone.

I’ve not made any crazy amount of requests with user id “1” - so wondering why would this happen to one user but not another?

Any advice much appreciated :pray:

Cheers,
Peter

1 Like
2

I have only user ID and I’m getting only 503 status all the time. It started about 2-3 hours ago.

1 Like
3

Thank you for letting me know @dbk03 - I’ve pulled out a lot of my hair over this today :sweat_smile:

4

Many of my users are getting 503 errors today, and there seems to be no rhyme or reason. The status page says everything is operational. Not true.

2 Likes
5

Thanks @sms - there was clearly an issue, had the status page reflected it I’d go sit in the sun in the garden, otherwise I spent digging into it like crazy - what a waste of a day :joy:

Btw things seem to be back to normal - no more 503, same for you?

6

Yep, no more errors. Seems to be resolved. I did report it…not sure if that helped. The last time I reported an API error to support, I got a reply weeks later, and they told me to clear my browser cache. :woman_facepalming: :laughing:

1 Like
7

Haha yeah happened to me, but the response came back months later saying to look at the status page (it was saying all is good) :face_with_peeking_eye:

8

We had this problem on Friday too, and the problem seems to be back this morning. It works fine when I use my user ident, but doesn’t work for some of the others users’ ident. Is there a partial outage at openai going on?
Using model gpt-3.5-turbo only, gpt-4 seems to work ok.

9

Yes, experiencing frequent outages on GPT 3.5 for the past 3 hours today

1 Like
10

Same here, the GPT 3.5 seems to be overloaded for chat completion

11

Keep getting continual 503 errors here too. Maybe 1 in 10 requests goes through normally.

We are sending extremely light volume.

12

Happened again this morning. Also sending light volume.

13

Same for me, maybe since they changed the prompts price the demand increased. To make it work I switch to gpt-3.5-turbo-16k but it’s more expensive

14

I’ve been noticing a lot of 503/500 errors also. I might be crazy but changing the “name” field on my request will toggle between a successful and error response.

Error 503:

{
    "model": "gpt-3.5-turbo",
    "messages": [
        {
            "role": "user",
            "content": "Hi."
        }
    ],
    "max_tokens": 800,
    "temperature": 0.8,
    "frequency_penalty": 0,
    "user": "lVJjDMAsYk5uo9seH7zmC4WJe933"
}

Successful (I’ve deleted the last character “3” on the “name”)

{
    "model": "gpt-3.5-turbo",
    "messages": [
        {
            "role": "user",
            "content": "Hi."
        }
    ],
    "max_tokens": 800,
    "temperature": 0.8,
    "frequency_penalty": 0,
    "user": "lVJjDMAsYk5uo9seH7zmC4WJe93"
}