TestLocalAdoptionSavesAndReloadsEligibleModel flakes under the race detector: its probe uses the 400ms default timeout #1132

Version / branch / commit

Seen on #886 at 433653b8, which has main at 4dc603f6 merged in. It failed in the Race Detector job on the first attempt, and a re-run of the same commit passed.

What happened

--- FAIL: TestLocalAdoptionSavesAndReloadsEligibleModel (0.74s)
    provider_adoption_test.go:57: expected one detected runtime: []

Cause

The test calls provideronboarding.DetectLocalRuntimes against its own httptest server without setting Timeout, so the probe gets defaultLocalProbeTimeout, which is 400ms. probeLocalRuntime treats a timeout as "not running" and drops the candidate without an error, so one slow loopback round trip under -race on a busy runner becomes an empty result. The run took 0.74s, which fits a probe that hit the deadline. Nothing in #886 touches this path, and the test is about adoption rather than probe timing.

I couldn't reproduce it locally. Ten runs under -race, and twenty more with every core busy, all passed on a 20-core Windows machine.

Fix

Pass a test-sized Timeout of a few seconds in the test's LocalDetectOptions. The production default stays at 400ms.