#Error 524: Timeout Issue with API Calls to Multiple Models

5 messages · Page 1 of 1 (latest)

faint nebula
#

Hello OpenRouter Support Team,

I hope you're doing well. I'm encountering a 524 error when making API requests to OpenRouter, where the server times out after a few minutes of processing. Here are the specific details:

  • Models Called: Claude 3.5 Sonnet, OpenAI4o, 4o-omni, and o1-omni
  • Request Rate: I’ve tested various request rates, from 10 requests per minute to 100 requests per minute.
  • Outcome: All requests fail after running for 2-3 minutes, with higher request counts failing faster. This suggests that the server is experiencing resource overload, not necessarily related to the volume of data but possibly the frequency of requests.

I’ve included the relevant portions of the logs below. Based on my testing, the error occurs consistently across multiple models and request rates, pointing to a possible issue on OpenRouter’s end.

Relevant Log Excerpts:

2024-10-09 23:56:06,342 - ERROR - API request failed for model openai/gpt-4o. Status: 524  
Cloudflare Ray ID: 8d01f3d6adbcbebe

2024-10-09 23:56:06,344 - ERROR - API request failed for model anthropic/claude-3.5-sonnet. Status: 524  
Cloudflare Ray ID: 8d01f3d6ade977a5

2024-10-09 23:56:07,933 - ERROR - API request failed for model anthropic/claude-3.5-sonnet. Status: 524  
Cloudflare Ray ID: 8d01f3e0ab44bef8

All requests are showing the same 524 timeout error across models, and my network and Cloudflare status indicate that everything is working on my side. Could you confirm if this is a server-side issue, and if so, are there any suggestions or optimizations I can implement to avoid these timeouts?

Thank you for your assistance, and I appreciate any guidance you can offer.

Cheers,
Xule

valid sorrel
#

cc @raven niche

jagged lichen
#

I get the same error sometimes, it's very annoying

Cannot cc @sam coz we're not friends 😦

jolly widget
#

Getting the same error as well with LLAMA 3.2 11 b

jagged lichen
#

It's an issue for all models, I guess. I added a retry mechanism that attempts 3 times in my backend when I encounter a 524 error, and it helps for now.