Module worker 

Blocking worker thread, request dispatch, and probe wiring.

Submodules:

• config: per-worker execution policy (WorkerConfig).
• guard: lifecycle guard, JIT guard wiring, inference outcomes.
• trt_retry: TRT JIT-OOM detection and workspace-halving retry.
• propagation: engine propagation broadcast drain.
• probe: probe inference and arena priming helpers.
• prewarm_strict: prewarm postcondition readiness gate.
• startup: model load, cache GC, TRT prewarm, readiness signal.
• run: request loop and idle lifecycle.
• dispatch: per-request dense/sparse/dual/probe/adaptive-warmup handling.
• logging: abandoned-request observability.

Re-exports

pub use config::WorkerConfig;

Modules

config 🔒

Per-worker execution policy shared across the pool.

dispatch 🔒

Per-request dispatch for dense, sparse, dual, probe, and adaptive warmup.

guard 🔒

Worker lifecycle guard, in-band JIT guard wiring, and inference outcomes.

logging 🔒

Observability helpers for abandoned in-flight requests.

prewarm_strict 🔒

Prewarm postcondition readiness gate (BGE_M3_PREWARM_STRICT).

probe 🔒

Probe inference helpers for startup measurement and arena priming.

propagation 🔒

Engine propagation broadcast drain and post-inference cache-miss signaling.

run 🔒

Blocking worker thread and request dispatch loop.

startup 🔒

Worker startup: model load, optional cache GC, TRT prewarm, readiness signal.

trt_retry 🔒

TRT JIT-OOM detection and single retry with halved workspace budget.