403 Forbidden triggered by prompt content #1459
Description
What Happened?
Hi Portkey Team,
We are encountering a 403 Forbidden error when sending specific prompts through the Portkey Gateway. The same prompt works when we reprocess the file or run it offline.
For a single file, we have two different sets of fields and we call two APIs using the same configuration. One works, while the other intermittently fails with a 403. However, when we reprocess the file, the previously failing request also succeeds. We did not face this issue with the same prompt previously, but recently we are receiving 403 errors continuously even after adding a retry mechanism.
We would like to understand why this behavior is occurring, so that we can take appropriate recovery actions. This is critical for improving the robustness and reliability of our application.
Incident Details:
- Trace ID: e0eb6a07-c0fe-49fd-9d22-eb64948b94b4
- Timestamp: 2025-11-28 05:21:05 UTC
- Status: 403 Forbidden
- Server Header: Cloudflare
We have attached:
- the failed request payload
- the successful request payload
- the corresponding error logs of failed request payload
- the original file
[failed_request_payload.ipynb](https://github.com/user-attachments/files/23816600/failed_request_payload.ipynb)
[file_403.pdf](https://github.com/user-attachments/files/23816601/file_403.pdf)
[success_request_payload.ipynb](https://github.com/user-attachments/files/23816599/success_request_payload.ipynb)
What Should Have Happened?
No response
Relevant Code Snippet
from portkey_ai import Portkey,PORTKEY_GATEWAY_URL
from openai import OpenAI
import instructor
import os
def get_portkey_instructor_client_openai_flavour() -> Any:
openai_client = OpenAI(base_url=PORTKEY_GATEWAY_URL,api_key="dummy-key")
return instructor.patch(openai_client)
llm_provider='google'
llm_config_name_mapping = {
"google":os.environ["PORTKEY_GOOGLE_CONFIG_ID"],
"openai":os.environ["PORTKEY_OPENAI_CONFIG_ID"],
}
config_name = llm_config_name_mapping[llm_provider]
client = get_portkey_instructor_client_openai_flavour()
metadata={'vdoc_trace_id':'8d11326e7cb10f705f9e655ff516f77a'}
response = client.chat.completions.create(
model=model,
messages=[{"role": "user", "content": content_items}],
max_retries=0,
extra_headers={
"x-portkey-config": config_name,
"x-portkey-metadata": json.dumps(metadata),
"x-portkey-api-key": os.environ["PORTKEY_API_KEY"]
},
temperature=0.0,
seed=10,
response_model=response_model
)
Your Twitter/LinkedIn
No response