[S1ro1]

Workers that are still loading during startup can be indexed under unknown and receive no model-filtered traffic until model discovery runs again. Refresh /v1/models on every runtime health-check cycle instead of every fifth cycle, and lower the default cycle interval from 60 to 20 seconds.

The six-line change updates the Rust config, Rust CLI, Python binding and Python CLI defaults, the documented default, and the model refresh interval. Existing filtering to healthy workers and deduplication by DP base URL remain in place. Explicit interval overrides still work; unhealthy workers still need to meet the existing health recovery threshold before model discovery runs.

Validation: git diff origin/main --check passed. Executed checks for the Python configuration default, CLI default, and an explicit CLI interval override; all passed. Rust build and tests were not run because this environment has no Rust toolchain.

---

Note

Low Risk
Default timing and background polling only; explicit health_check_interval_secs overrides still apply and routing logic is unchanged aside from fresher model indexes.

Overview
Speeds up worker model discovery so workers are not stuck under unknown and excluded from model-filtered routing while they finish loading.

Default runtime health-check interval is lowered from 60s to 20s across Rust config/CLI, Python RouterArgs/docs, and PyO3 binding defaults. In the registry health checker, MODEL_REFRESH_INTERVAL goes from 5 to 1, so /v1/models refresh runs on every health-check cycle instead of every fifth cycle (with healthy-worker filtering and DP base-URL dedup unchanged).

^{Reviewed by Cursor Bugbot for commit 6bb5701. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Change default health-check interval from 60 to 20 seconds

Updates the default for health_check_interval_secs from 60 to 20 across the Python dataclass, PyO3 constructor, Rust HealthCheckConfig default, and CLI args. Also updates the Router class docs. The model-refresh interval inside WorkerRegistry.start_health_checker changes from 5 seconds to 1 second.

• Behavioral Change: routers built without an explicit interval now run the health check every 20 seconds; the health-check task refreshes worker models every 1 second — see worker_registry.rs and router_args.py.

📊 Macroscope summarized 6bb5701. 6 files reviewed, 1 issue evaluated, 0 issues filtered, 1 comment posted

🗂️ Filtered Issues

[macroscopeapp]

🟠 High core/worker_registry.rs:477

Health checks are continuously delayed when multiple /v1/models requests stall, so failed workers can remain marked healthy and continue receiving traffic. With MODEL_REFRESH_INTERVAL set to 1, every health-check cycle performs sequential discovery requests, each potentially taking five seconds; restore the less frequent refresh interval (or decouple discovery from health checks).

🚀 Reply "fix it for me" or copy this AI Prompt for your agent:

In file @src/core/worker_registry.rs around line 477:

Health checks are continuously delayed when multiple `/v1/models` requests stall, so failed workers can remain marked healthy and continue receiving traffic. With `MODEL_REFRESH_INTERVAL` set to `1`, every health-check cycle performs sequential discovery requests, each potentially taking five seconds; restore the less frequent refresh interval (or decouple discovery from health checks).

[macroscopeapp]

Approvability

Verdict: Not approved

Macroscope's review found this PR not approvable — The PR changes production health-check and model-discovery cadence, causing every default health cycle to perform potentially slow sequential /v1/models requests. That coupling can affect failure detection and traffic routing under stalled workers, so the runtime impact needs careful validation.

Not approved because:

• 1 blocking correctness issue found at or above your repo's Minimum Blocking Severity

^{Adjust the Minimum Blocking Severity for this repo — including turning it Off — in Settings. You can add or adjust custom eligibility rules. Learn more.}