Replaces per-call thread spawning for bounded filesystem probes with a reusable pool of up to eight helpers, retiring timed-out helpers and replacing them on demand within the existing 64-thread cap.
Reused helpers reset helper-local directory, mutation-admission, read, canonicalization, and grep-audit state before each dispatch and restore it on unwind.
A dedicated factory thread keeps worker creation off timed admission paths.
Health is non-blocking: an unused pool reports not_started, and a contended pool reports unavailable rather than waiting for the state lock.
The health report includes filesystem_helpers with pool limits, abandonment, replacements, and admission counters.
A 900-call read/grep/edit workload drops helper spawns from 5400 to 1 and process CPU per dispatch from 315.6 to 270.0 µs.
Adds regression tests for reuse, timed-out replacement, cap recovery, context reset, unwind restoration, and non-blocking health.
Confidence Score: 5/5
The PR appears safe to merge, with no outstanding actionable findings.
Summary
Reuses filesystem helpers within an eight-worker pool and a 64-thread cap. Timed-out helpers leave the reusable pool and can be replaced.
Filesystem probes reuse helper threads and retire timed-out workers.
Filesystem health reports show the probe pool’s status and limits.
Diagram
%%{init: {'theme': 'neutral'}}%%
flowchart TD
Request[Filesystem probe] --> Pool[Acquire helper before deadline]
Pool --> Worker[Run job with clean local state]
Worker --> Success[Return helper to pool]
Worker --> Timeout[Retire timed-out helper]
Timeout --> Replacement[Create replacement within thread cap]
Health[Health read] --> Snapshot[Report ready, not_started, or unavailable]
