Cannot complete a single task.

[Savalinn]

Type: Bug

"Sorry, the upstream model provider is currently experiencing high demand. Please try again later or consider switching models."

I accept that your service is busy, but why interrupt an ongoing flow? If your load is too high block it before at the beginning, not somewhere in the middle of a process after 3-5 mins.
Please reconsider your "busy-interruption-strategy"

Extension version: 0.32.5
VS Code version: Code 1.105.1 (7d842fb, 2025-10-14T22:33:36.618Z)
OS version: Windows_NT x64 10.0.26200
Modes:
Remote OS version: Linux x64 6.1.0-39-amd64

System Info

Item	Value
CPUs	AMD Ryzen 7 5800X 8-Core Processor (16 x 3793)
GPU Status	2d_canvas: enabled direct_rendering_display_compositor: disabled_off_ok gpu_compositing: enabled multiple_raster_threads: enabled_on opengl: enabled_on rasterization: enabled raw_draw: disabled_off_ok skia_graphite: disabled_off trees_in_viz: disabled_off video_decode: enabled video_encode: enabled vulkan: disabled_off webgl: enabled webgl2: enabled webgpu: enabled webnn: disabled_off
Load (avg)	undefined
Memory (System)	31.93GB (17.19GB free)
Process Argv	--crash-reporter-id 7562b6b7-5191-421a-834e-63bb0a32594d
Screen Reader	no
VM	27%

Item	Value
Remote	SSH: devsrv
OS	Linux x64 6.1.0-39-amd64
CPUs	Intel(R) Core(TM) i9-9900K CPU @ 3.60GHz (8 x 0)
Memory (System)	7.75GB (3.97GB free)
VM	0%

A/B Experiments

vsliv368:30146709
binariesv615:30325510
nativeloc1:31344060
dwcopilot:31170013
dwoutputs:31242946
copilot_t_ci:31333650
e5gg6876:31282496
pythonrdcb7:31342333
6518g693:31334701
aj953862:31281341
nes-set-on:31351930
6abeh943:31336334
envsactivate1:31353494
aa_t:31379598
c12345:31413568
gaj49834:31404438
cloudbuttont:31379625
todos-1:31405332
qwen_all_req:31402696
treatment_gpt5applypatchexclusively:31412086
3efgi100_wstrepl:31403338
trigger-command-fix:31379601
auto_model_enabled:31385282
use-responses-api:31390855
sonnet_45_3e700918:31415506
ddidtcf:31399634
je187915:31401257
v2prompt_9f72f763:31415580
afag3459:31405348
ec5jj548:31418076
copilot-nes-oct-t:31418199
ff8f5884:31411053

[vs-code-engineering]

Thanks for creating this issue! It looks like you may be using an old version of VS Code, the latest stable release is 1.106.0. Please try upgrading to the latest version and checking whether this issue remains.

Happy Coding!

[lramos15]

Have you tried switching the model? These providers are upstream from us and can ocassionally get overwhelmed. They have no concept of what an interaction is so the best thing we can offer the user is to switch to a new model

[Savalinn]

Thanks for the reply, but I think there is a misunderstanding.

My issue is not about the model provider being busy.
My issue is about how the VS Code extension handles this situation.

Right now the extension starts the request, begins the streaming, and then after 3–5 minutes it suddenly stops with the error “the upstream model provider is experiencing high demand”. This means the whole interaction is lost in the middle.

The problem is not the provider load itself.
The problem is that the extension detects the overload too late and interrupts an ongoing generation. If the system is already under high load, the extension should fail before starting the answer, not during the stream.

Please check the timeout/error-handling logic. Early failure or retry would give a much better user experience than interrupting a long running answer halfway.

[lramos15]

If the system is already under high load, the extension should fail before starting the answer, not during the stream.

This isn't information we receive. We're working to implement auto mode which will alleviate some of this by working with our providers to route us to less busy GPU pools, but I don't see much we can currently do to alleviate this issue. But definitely makes sense!