[cdalar]

Why

Firecracker chooses the kernel at boot, and onctl had no way to boot an existing VM again:

• resume restores a snapshot, with the old kernel and memory inside it;
• create over a dead record (e.g. after a host reboot) wipes the VM directory, including its rootfs.

So moving a VM to a new kernel, or recovering one whose VMM died, meant losing its disk. This came up when boxctl-vms started booting boxes on its own kernel (cdalar/boxctl-vms#99, #100): existing boxes, including multi-node k3s clusters, could only get it by being recreated.

What

• cloud.Restarter (optional interface, like ImageLister) + RestartOptions{KernelImage}.
• ProviderFC.Restart: best-effort guest filesystem sync, stop the VMM if it's alive, recreate the TAP device, and boot a fresh VMM against the VM's existing rootfs (and cache disk) with the same IP, MAC, vCPUs, memory and kernel command line.
  • --kernel-image swaps the kernel and is saved in the metadata, so later restarts keep it.
  • Works for running, dead (host reboot) and paused VMs (the snapshot is discarded: a cold boot can't resume it, and it would bring back the old kernel).
  • A missing kernel/rootfs fails before the VM is touched; a failed boot records the VM as dead instead of leaving stale "running" metadata.
• Deploy and Restart share kernelArgs(), so a restarted VM gets exactly the network settings it was deployed with.
• onctl restart <name> [--kernel-image <vmlinux>] [-f] (fc only; other providers get a clear "not supported").
• TESTING-fc.md: manual test line.

Testing

• New unit tests: running VM + new kernel (same disk/IP/MAC/cmdline, new kernel persisted), default kernel kept, dead VM, paused VM (snapshot removed, cold boot not restore), missing kernel leaves the VM running, boot failure marks dead, not found; command flags/args.
• go test ./pkg/cloud/ ./internal/... pass, go vet, golangci-lint clean.
• cmd: TestReadConfig_NoConfigDirectory and TestCreateFlagsBindToViper fail locally on main as well (unrelated; they seem to depend on the local ~/.onctl).
• Not yet run against real Firecracker. Suggested on aimax: sudo ONCTL_CLOUD=fc onctl -c /root/.onctl/onctl.yaml restart <vm> -f --kernel-image /opt/fc/images/boxctl-kernel.vmlinux, then uname -r / check the disk content survived.

Follow-up (not in this PR)

boxctl-vms could expose this (e.g. "restart box" / automatic kernel upgrade on restart), since its agent already dispatches onctl commands.

🤖 Generated with Claude Code

https://claude.ai/code/session_01WubXQDJdsuKoncpCSToz93

[codecov]

Codecov Report

❌ Patch coverage is 65.71429% with 36 lines in your changes missing coverage. Please review.
✅ Project coverage is 34.54%. Comparing base (9d31e7a) to head (b9cabbf).

Files with missing lines	Patch %	Lines
pkg/cloud/fc.go	76.47%	20 Missing ⚠️
cmd/restart.go	20.00%	16 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1140      +/-   ##
==========================================
+ Coverage   33.99%   34.54%   +0.54%     
==========================================
  Files          50       51       +1     
  Lines        5509     5610     +101     
==========================================
+ Hits         1873     1938      +65     
- Misses       3636     3672      +36     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.