vscode plugin experiences (timeout?) errors in ask mode

[vugenti]

Working with VSCode plugin, using "AGENT" or "ASK" mode and providing several files, after a few turns in the conversation and with a relatively small amount of context used, VS Code often "gives up" while lemonade server is still processing the request (quite often at 99% complete - but depends on the complexity of the prompt and what other applications are running in the background) and responds with an error. I am guessing it is some form of timeout as the message is indicating "terminated at Fetch.onAborted" - I looked for but could not find any configuration for timeout settings - is this within lemonade and/or is there any way for the plugin to influence or improve the experience in this situation:

Sorry, your request failed. Please try again.

Copilot Request id: [redacted]

Reason: terminated: TypeError: terminated at Fetch.onAborted (node:internal/deps/undici/undici:11322:53) at Fetch.emit (node:events:519:28) at Fetch.terminate (node:internal/deps/undici/undici:10480:14) at Object.onError (node:internal/deps/undici/undici:11445:38) at Request.onError (c:\Users\redacted\AppData\Local\Programs\Microsoft VS Code Insiders\37f3964564\resources\app\node_modules\undici\lib\core\request.js:316:27) at Object.errorRequest (c:\Users\redacted\AppData\Local\Programs\Microsoft VS Code Insiders\37f3964564\resources\app\node_modules\undici\lib\core\util.js:798:13) at Socket.onHttpSocketClose (c:\Users\redacted\AppData\Local\Programs\Microsoft VS Code Insiders\37f3964564\resources\app\node_modules\undici\lib\dispatcher\client-h1.js:928:10) at Socket.emit (node:events:531:35) at TCP. (node:net:346:12) at TCP.callbackTrampoline (node:internal/async_hooks:130:17)

This is the actual context used at the time the error occurred:

Hardware - AMD Ryzen AI Max 395+ w/ 128GB RAM

VS Code Insiders version info:

Version: 1.110.0-insider (user setup)
Commit: 37f39645643438347f1b5077abdaa96e7297fa7c
Date: 2026-02-27T03:12:42.551Z
Electron: 39.6.1
ElectronBuildId: 13369494
Chromium: 142.0.7444.265
Node.js: 22.22.0
V8: 14.2.231.22-electron.0
OS: Windows_NT x64 10.0.26200

[vugenti]

lemonade server logs from the error (occurs immediately after the prompt finishes processing, before inference begins - not the same time when the VS Code error is displayed)

[HttpClient ERROR] CURL error: Failed writing received data to disk/application
[WrappedServer ERROR] Streaming request failed: CURL error: Failed writing received data to disk/application
[Server] POST /api/v1/chat/completions - 200
srv log_server_r: done request: POST /v1/chat/completions 127.0.0.1 200
srv stop: cancel task, id_task = 10716
slot release: id 3 | task 10716 | stop processing: n_tokens = 105858, truncated = 0
srv update_slots: all slots are idle

My current workaround is to watch lemonade server logs in the background and wait for the prompt to complete, then in VS Code I click "RESTORE CHECKPOINT" to last turn in the conversation and re-add the exact same files for context, then resubmit. If the prompt is served from cache and returns immediately, the session is able to continue. This seems extremely effective, but if I slip up and attach the wrong files, it throws out the cached response and starts over.

[vugenti]

Also noticing this in the lemonade server status bar (TTFT looks suspicious to me):

[vugenti]

My current workaround is to watch lemonade server logs in the background and wait for the prompt to complete, then in VS Code I click "RESTORE CHECKPOINT" to last turn in the conversation and re-add the exact same files for context, then resubmit. If the prompt is served from cache and returns immediately, the session is able to continue. This seems extremely effective, but if I slip up and attach the wrong files, it throws out the cached response and starts over.

Unfortunately, in Agent mode when the agent performs multi-prompt operations such as reviewing its own work, VS Code does not create a checkpoint and ends up with half implemented changes and no way to restore the context to continue from the failed prompt, so this workaround is only for Ask mode.

[james-martinez]

Getting same issues on large requests using open ai compatible api and ollama api

[Server] POST /api/v1/chat/completions - Streaming
[HttpClient ERROR] CURL error: Failed writing received data to disk/application
[WrappedServer ERROR] Streaming request failed: CURL error: Failed writing received data to disk/application

Switched to ollama api and still fails but completes the first post request at 62k prompt tokens. No errors in log

Switched to straight llama-server.exe and its working . Copied and pasted from server log. That's my current work around.

Will provide more info on "disk" error using resource manager.

Might be related to lemonade-sdk/lemonade#572 (comment)

Seems to be the static timeout 300L in proxy and wrapper? HttpClient::post

makes sense if im getting 200 prompt tokens per second. around 60k it breaks. ~5mins, 300s

[jeremyfowers]

Hi folks, the VS Code plugin is now open source, so I moved the issue to https://github.com/lemonade-sdk/lemonade-vscode

FYI we need a community maintainer for this codebase right now, and @danielholanda (the solo maintainer) is looking members of the community who want to step up.