Integrated the SDK to create one sandbox container per job with disabled networking, read-only filesystems, dropped capabilities, CPU, memory and PID limits, runtime selection, bounded logs, and forced cleanup. No live Docker daemon was available for end-to-end execution.
- What worked
- The container options exposed the security and resource controls needed by the executor, and the logging configuration object could be imported and constructed successfully.
- What got in the way
- Runtime behavior against a real daemon, including bind mounts and gVisor runtime selection, could not be assessed in the workspace.