# verae-keep

Host keep-alive for Verae workers. Restarts a unit that crashed **unless** the operator console paused or stopped it.

**Forgejo:** https://git.georgelambert.org/marchon/verae-keep

Two processes:

| Process | Port | Job |
|---------|------|-----|
| `node src/keep.js` | `:3860` | Probe units, spawn if intent is `run` |
| `node src/watch.js` | `:3861` | Restart keep if `/health` fails |
| `scripts/guard.sh` | — | Restart watch if watch exits |

Intent is `run` | `pause` | `stop`. Sources, newest wins:

1. `POST /intent/:id` (admin / API)
2. `~/verae-fleet-runtime/data/<instance>/intent.json` (fleet writes this on start/pause/stop)
3. live `FLEET_URL/api/status` when set
4. unit `defaultIntent` (`run`)

Pause: leave the process down or POST `/pause`. Stop: SIGTERM and do not spawn. Run: spawn if health fails.

## NS1 (70.88.205.138)

```bash
KEEP_UNITS=$HOME/verae-keep/units.ns1.json KEEP_STATE=$HOME/verae-keep/data \
  nohup bash scripts/guard.sh >/tmp/verae-keep-guard.out 2>&1 &
```

Or user systemd (survives reboot; do not also `nohup guard.sh`):

```bash
bash scripts/install-systemd.sh
systemctl --user start verae-keep-guard
```

Proxmox guest extra copies: `KEEP_UNITS=units.px-worker.json`.

Units: NATS (observe only), job-poller, webhook-deliver, archive-aggregator, archive-worm ×3, tree-node ×3.

## API

- `GET /health` `GET /status`
- `POST /intent/:id` `{"state":"pause|stop|run"}`
- `POST /reconcile`