fix: add timeout to chat control data fetch request (fixes #317337)

[vs-code-engineering]

🔧 Error Fix

Summary

The _fetchChatControlData function in the chat language models service makes a periodic network request (every 5 minutes) to fetch chat control data without a request timeout. When the network is slow or unresponsive, the native fetch call hangs indefinitely, causing the CPU profiler to sample it and generate a PerfSampleError: by <> in fetch that reaches error telemetry.

This is a new anomaly in stable 1.120.0, affecting 3,958 users with 6,332 hits across all platforms (Mac, Windows, Linux).

Fixes #317337

Recommended reviewer: @lramos15

Culprit Commit

Commit range narrowing was inconclusive per the regression scan. The _fetchChatControlData function existed prior to 1.120.0, but the absence of a timeout has always been latent. The anomaly likely appeared due to changes in the performance sampling infrastructure or increased server latency in 1.120.0.

Code Flow

sequenceDiagram
    participant LM as languageModels.ts
    participant RS as requestService
    participant RI as requestImpl.ts
    participant Net as Network/fetch
    participant Prof as Profiler

    LM->>LM: _refreshChatControlData() [every 5 min]
    LM->>RS: request({url, callSite, NO timeout})
    RS->>RI: logAndRequest -> request()
    RI->>Net: fetch(url, {signal})
    Note over Net: Network slow/unresponsive
    Prof->>Prof: CPU sample detects fetch on stack
    Prof->>Prof: Creates PerfSampleError
    Prof->>Prof: errorHandler.onUnexpectedError()
    Note over Prof: Error reaches telemetry

Loading

Affected Files

File	Role
src/vs/workbench/contrib/chat/common/languageModels.ts	Producer of the unbounded fetch call
src/vs/base/parts/request/common/requestImpl.ts	Crash site — where fetch executes (not modified)
src/vs/platform/profiling/common/profilingTelemetrySpec.ts	PerfSampleError generation (not modified)

Repro Steps

1. Open VS Code 1.120.0 on any platform
2. Ensure network conditions are degraded (slow DNS, throttled connection, or unresponsive chat control endpoint)
3. Wait for the 5-minute _refreshChatControlData cycle to fire
4. The fetch call hangs; the CPU profiler samples it and emits a PerfSampleError
5. Error appears in telemetry as unhandlederror-PerfSampleError: by <> in fetch

How the Fix Works

Chosen approach (src/vs/workbench/contrib/chat/common/languageModels.ts):

Added timeout: 30_000 (30 seconds) to the request options passed to this._requestService.request() in _fetchChatControlData. This causes requestImpl.ts to use AbortSignal.timeout(30000) which aborts the fetch after 30 seconds. The abort throws a TimeoutError which is caught by the existing try/catch in _fetchChatControlData (line 1929), logged as a warning, and the function returns gracefully. The next retry happens in 5 minutes via _refreshChatControlData.

This fixes the issue at the data producer level — the unbounded request that causes the profiler to flag the slow fetch. The fix prevents the fetch from hanging long enough to be profiled as a performance problem, without silencing any errors or removing telemetry logging.

Recommended Owner

@lramos15 — most active recent contributor to languageModels.ts with model management and picker changes.

Generated by errors-fix · ● 48.8M · ◷

[Copilot]

Copilot encountered an error and was unable to review this pull request. You can try again by re-requesting a review.

[Copilot]

Copilot encountered an error and was unable to review this pull request. You can try again by re-requesting a review.

[Copilot]

Copilot encountered an error and was unable to review this pull request. You can try again by re-requesting a review.