[lunarwhite]

Summary

Keep the Docker supervisor healthcheck's 250 ms start_interval for fast readiness detection and raise the steady-state interval from 250 ms to 30 s, as proposed in the triage.

Related Issue

Closes #4729

Changes

• Give the start interval its own constant (250 ms) and set SUPERVISOR_HEALTH_INTERVAL_NS to 30 s. Timeout, retries, and the 60 s start period are unchanged.
• Move the HealthConfig into docker_supervisor_healthcheck() and pin all three timing values in a unit test.
• Note: Docker uses the start interval only within the 60 s start period, so a supervisor that becomes ready later (for example after a policy repair) is detected up to 30 s later. Public Ready comes from the supervisor session and is unaffected.

Testing

• Checks appropriate to the affected code and behavior pass
• Unit tests added/updated (if applicable)
• E2E tests added/updated (if applicable)

Additionally, one idle sandbox, Docker 29.7.2 on WSL2, gateway built from main vs this branch:

	main	This PR
Health probes per minute	185	2
Supervisor container CPU	16.3%	0.4%
dockerd CPU	4.7%	0.3%
sandbox start (3 runs)	1.7–2.8 s	1.8–2.7 s
Policy repaired 91 s into startup: to Ready / to Docker healthy	1.4 s / 1.2 s	1.0 s / 29.1 s

Checklist

• Follows Conventional Commits
• Commits are signed off (DCO)
• Architecture docs updated (if applicable)

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.