[Bug]: Receiving the response "Request failed" in chat with Copilot integration (intermittent)

[wright-io]

Describe the issue

Occasionally, I receive a "request failed" response from Copilot chat with CodeGate.

Steps to Reproduce

• Install CodeGate and configure to work with Copilot as per docs
• Issue chat request in Copilot
• Occasionally receive the "request failed" response

Operating System

MacOS (Arm)

IDE and Version

VS Code 1.96.0

Extension and Version

Copilot extension 1.252.0

Provider

OpenAI

Model

GPT 4o

Logs

2024-12-17T18:50:42.017Z [error    ] Pipeline processing error: Expecting value: line 1 column 1 (char 0) module=pipeline pathname=/app/src/codegate/providers/copilot/pipeline.py
2024-12-17T18:51:35.017Z [error    ] Pipeline processing error: Expecting value: line 1 column 1 (char 0) module=pipeline pathname=/app/src/codegate/providers/copilot/pipeline.py

Additional Context

Error response from Copilot:

Sorry, your request failed. Please try again. Request id: 6b142432-5ffc-43d1-8b4b-5230d239ba7a

[lukehinds]

@wright-io do you recall if this was several or a significantly large file? I have seen this before and I believe I have a fix

[jhrozek]

I've been trying to hammer the chat to reproduce the issue and I've been intermittently able to.
I don't know what is the root cause yet, just saying it's not just you :-)

[lukehinds]

@jhrozek load up lots of big files (500+) lines, that seems to do it. I might have a fix for this I can push, just need to finish off some TLS rework code.

[wright-io]

@wright-io do you recall if this was several or a significantly large file? I have seen this before and I believe I have a fix

I don't believe I had any files in the context. But there was a lot of info in the chat itself. I had it generating some code and iterating that multiple times in the same chat and then I sent an example of a file I wanted it to use as a model.

It was certainly hundreds of lines.

[jhrozek]

@jhrozek load up lots of big files (500+) lines, that seems to do it. I might have a fix for this I can push, just need to finish off some TLS rework code.

I was able to trigger this without a context at all, but instead by asking a lot of nonsensical questions in quick succession. This might actually be the same as big files as context as our system prompts are quite big and make for a large message.