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
feat: VM-level container logs, named volume tmpfs, Docker health semantics
Three fixes that get both postgres services in gpu-cli-two passing health checks:
1. **VM-level container logs**: Bind-mount `/run/vz-oci/logs/{cid}/` into
containers at `/var/log/vz-oci/` so stdout/stderr logs survive container
death. Added `exec_host()` on Runtime for VM-level command execution
without nsenter. `logs()` now reads from VM level, working even when
the container's init process has exited.
2. **Named volumes use tmpfs**: VirtioFS doesn't support chown/chmod from
the guest, which breaks containers like postgres that change data dir
ownership. Named volumes now use tmpfs inside the VM (matching Docker
Desktop behavior) instead of VirtioFS host-backed storage.
3. **Docker health check semantics**: When retries are exhausted, container
stays Running (unhealthy) instead of being killed and recreated. Health
checks continue indefinitely; a future pass promotes to healthy.
Also includes exec_with_output on ContainerRuntime trait, OCI service
implementation in guest agent, and shared VM volume mount support.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
0 commit comments