When the operator clicks Start vetting and the host is heartbeating,
the heartbeat response now carries cmd=reboot_for_vetting + run_id.
The handler drives the Queued → WaitingWoL transition via the existing
state machine, so a benign race with the 2s dispatcher poll is refused
by the state machine (not double-dispatched). WaitingWoL retries for
10 minutes to cover a crashed-mid-reboot case, then falls back to
operator action.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
vetting-agent gains a `host` subcommand that runs as a systemd service
installed by the quick-register one-liner, POSTing every 30s to
/api/v1/hosts/{mac}/heartbeat so the dashboard tile shows "online" or
"Nm ago" without waiting on WoL. Ships dormant client code for the
Phase 2 reboot_for_vetting command so the server can flip it on later
without a binary redeploy.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Two bugs compounded on Proxmox hosts: primary_iface walked
`ip link show` and picked the physical NIC (e.g. enp1s0), which has
no IPv4 on Proxmox because the address lives on vmbr0. Even if vmbr0
had been picked, the kernel reports its broadcast as 0.0.0.0, so the
script fell all the way back to 255.255.255.255.
Now we prefer the default-route interface (vmbr0 on Proxmox, eno1 on
bare metal) and, when `ip` doesn't surface a usable `brd`, compute
the broadcast from the inet CIDR instead of giving up.
Operator pastes `curl -fsSL $ORCH/register/quick.sh | sudo bash` on the
target host (pre-wipe). The script probes MAC + CPU/RAM/disks/NICs/GPUs,
emits an expected-spec YAML, and POSTs to a new LAN-trusted JSON
endpoint /api/v1/hosts. The register page shows the command prefilled
with the orchestrator URL; the manual form moves into a collapsible
"Register manually" disclosure.
Can't log in from a fresh LXC deploy, and the service is LAN-only by
design. Rip out the whole bcrypt-password / signed-cookie session
layer: internal/auth, login templates, gen-admin-password binary +
Makefile targets, auth config block, login/logout routes and the
RequireSession middleware wrap. Agent bearer-token auth on
/api/v1/runs/{id}/* is untouched.
Operators who want a password can front the service with a reverse
proxy — noted in README and docs/operations.md.
Service was crashing on every boot because vetting.example.yaml uses
./var/... relative paths that resolve to / under ProtectSystem=strict.
Ship a separate vetting.production.yaml with absolute /var/lib/vetting
+ /var/log/vetting paths that match the unit's ReadWritePaths, and
have install.sh copy that one. Also move StartLimit* keys into [Unit]
to silence the 'Unknown key' warning on modern systemd.
proxmox-install.sh + install.sh left operators with no way to
generate the bcrypt hash on the LXC — 'vetting gen-admin-password'
was suggested in the post-install message but the binary has no
subcommands. Cross-build gen-admin-password-linux-amd64 during the
one-liner flow and drop it into /usr/local/bin.
deploy/proxmox-install.sh bootstraps a fresh LXC end-to-end: apt
prereqs, Go toolchain (if missing), git clone, build, then hands off
to deploy/install.sh. README documents the curl|bash invocation.
josh