Reduce cbox sandbox startup latency by parallelizing health checks

Problem

Sandbox bootstrap health checks (python, pre-commit, uv, git_auth) ran sequentially via list comprehension, making worst-case wall time 4×15s=60s. Replaced with ThreadPoolExecutor to run all probes concurrently; worst case now ~15s.

_run_container_health_checks() in cbox/cli.py executed 4 independent docker exec probes one after another. Each has a 15s timeout. When multiple probes are slow (common in cold-start sandboxes), startup delay compounds.

Context

• Primary file: libs/cbox/cbox/cli.py — _run_container_health_checks() (lines 1255-1339)
• Types: libs/cbox/cbox/health.py — HealthCheckResult dataclass
• Tests: libs/cbox/test_sandbox_start.py — 7 health-check unit tests
• Constraint: _run_probe is a pure function (no shared mutable state), making it inherently thread-safe.
• Import: concurrent.futures.ThreadPoolExecutor at module level (line 16 of cli.py).

Possible Solutions

1. ThreadPoolExecutor — standard library, minimal change, futures preserve insertion order. ✓
2. asyncio.gather — would require converting the entire call chain to async. Overkill.
3. multiprocessing.Pool — unnecessary process overhead for I/O-bound subprocess calls.

Plan

Option 1: concurrent.futures.ThreadPoolExecutor with max_workers=len(checks). Submit all probes, collect results via f.result() in definition order. Three-line change in _run_container_health_checks().

Implementation Progress

• Run health checks (python, pre-commit, uv, git_auth) concurrently.
• Preserve deterministic output ordering in CLI rendering.
• Keep per-check timeout and remediation behavior unchanged.

• Add/adjust tests to validate concurrency-safe behavior.

• Worst-case health check wall-clock time is near single-check timeout, not sum of all check timeouts.

• Existing warning panel output remains stable and actionable.

• cbox tests pass.

• Source issue: repeated 6-12 minute reviews plus additional startup delay from sequential checks.

• Changed the return line from [_run_probe(...) for ... in checks] to ThreadPoolExecutor.submit() + [f.result() for f in futures].

• Order preserved because futures list matches checks list order.
• Per-check 15s timeout unchanged (each thread still calls subprocess.run(timeout=15)).
• _run_probe was already a pure function with no shared mutable state — thread-safe by design.
• Added two tests: test_container_health_checks_run_concurrently (timing) and test_container_health_checks_preserve_order (deterministic output).

Review Feedback

Round 1 — cbox review (opus), 2026-03-08

• Verdict: APPROVED
• MEDIUM-1: Move concurrent.futures import to module level → Fixed.

• MEDIUM-2: Order test doesn't exercise out-of-order completion → Fixed (added staggered delays).

• [x] Review cleared