Description
Currently, sandbox environments behave like a black box. When processes die (e.g., OOM, crash), users cannot see resource usage or termination reasons. This makes it hard to distinguish between application bugs and infrastructure issues.

Proposal

• Expose per-sandbox metrics (CPU, memory, disk, I/O) via API and dashboard.
• Surface process termination reasons (e.g., OOM kill, manual stop, internal crash).
• Enable integration with external observability tools (e.g., OTEL collector endpoint configuration).
• (Optional) Provide a built-in lightweight supervisor to restart main processes when killed.

Acceptance Criteria

• API returns real-time and historical CPU/memory/disk usage for each sandbox.
• Sandbox events include termination reason with timestamp.
• Dashboard displays metrics graphs and termination events.
• Configurable OTEL collector endpoint supported for streaming metrics/logs.
• (Optional) Supervisor process can be toggled per sandbox for auto-restarts.

Impact

• Users can self-diagnose issues like OOM without needing Daytona support.
• Faster debugging and reduced downtime for production-like workloads.
• Enables external monitoring and alerting pipelines (Datadog, Grafana, etc.).
• Improves reliability and trust.