You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Extend the current Runtime observation API and Core Web Dashboard from Docker to the configured single-host microsandbox provider without adding a second public contract or lifecycle authority.
Context
Follow-up to #19 and PR #15. The pinned microsandbox Go SDK v0.7.2 exposes point-in-time metrics through SandboxHandle.Metrics, including cumulative vCPU time, memory usage/limit and uptime. Core must continue to invoke the SDK only through the existing one-shot Linux helper.
Scope
implement runtimeobs.Source on the microsandbox provider
add a read-only helper metrics operation with exact allocation and compute-incarnation verification
normalize vCPU time to cumulative CPU seconds, configured vCPU count to capacity, SDK memory usage/limit, and uptime to compute start time
map stopped/suspended compute and disabled/unavailable SDK metrics to existing safe observation states
register the selected single managed provider as the observation source after the current main/provider configuration merge
update provider-neutral design/deployment documentation and focused tests
Boundaries
no new Dashboard, public endpoint, migration, history store or lifecycle action
no token duplication; Dashboard keeps joining canonical Session usage
no use of instantaneous SDK CPUPercent for the cumulative CPU contract
disk, network, host RSS and overlay metrics remain future public-contract work
no provider-native names, IDs, labels, paths or raw SDK errors in public responses
no sampling of suspended snapshots as zero resource usage
Acceptance
an owned running current compute returns the same normalized CPU/memory fields as Docker
restored generations are sampled by the persisted exact compute identity, never by display-name search
Activity