fleet_man

A lightweight Python library for running and monitoring scripts on Raspberry Pis from a browser. Each Pi hosts its own control panel with live log streaming. An optional fleet coordinator aggregates all Pis into a single dashboard.

Built on FastAPI + WebSockets. No external message broker. No database. Each Pi is fully autonomous.

---

install

pip install 'uvicorn[standard]' fastapi httpx

---

quick start — single Pi

# app.py
import sys
import uvicorn
from pathlib import Path
from process_manager import ProcessManager, ScriptSpec

PY = sys.executable

scripts = {
    "CAN-M":  ScriptSpec("CAN Manager",  [PY, "gc_monitor/run.py"]),
    "backup": ScriptSpec("backup",        ["bash", "jobs/backup.sh"]),
    "server": ScriptSpec("api server",    [PY, "jobs/server.py"]),
}

mgr = ProcessManager(
    name="pi-kitchen",
    scripts=scripts,
    log_dir=Path("/var/log/jobs"),
    port=5001,
)

app = mgr.create_app()

if __name__ == "__main__":
    uvicorn.run(app, host="0.0.0.0", port=5001)

Open http://<pi-ip>:5001 in your browser.

---

ProcessManager

ProcessManager(
    scripts,                          # dict[str, ScriptSpec]  — required
    name="pi",                        # node name shown in UI
    log_dir=Path("/tmp/pm_logs"),     # where .log files are written
    coordinator_url=None,             # optional fleet coordinator URL
    host=None,                        # LAN IP advertised to coordinator (auto-detected)
    port=8000,                        # port this Pi listens on
)

ScriptSpec

ScriptSpec(label, cmd)

arg	type	description
label	str	display name shown in the UI
cmd	list[str] or callable	command to run

cmd can be a callable — it is invoked at start time, so it can pull config from disk or environment dynamically:

def build_cmd():
    config = load_config()
    return ["python", "worker.py", "--env", config["env"]]

ScriptSpec("worker", build_cmd)

---

fleet coordinator

The coordinator runs on any central machine (laptop, NAS, another Pi). It polls each registered Pi, tracks online/offline status, and serves a unified fleet dashboard where you can start and stop scripts across all nodes.

# coordinator.py
import uvicorn
from coordinator import create_coordinator_app

app = create_coordinator_app(fleet_name="home-lab")

if __name__ == "__main__":
    uvicorn.run(app, host="0.0.0.0", port=9000)

Open http://<coordinator-ip>:9000.

auto-registration

Pass coordinator_url to ProcessManager and each Pi registers itself on startup and re-registers every 30 seconds — no manual steps needed. Boot order doesn't matter; Pis retry until the coordinator is reachable.

mgr = ProcessManager(
    name="pi-kitchen",
    scripts=scripts,
    coordinator_url="http://192.168.1.10:9000",
    port=5001,
)

You'll see confirmation in the Pi's console:

[pm] registered with coordinator at http://192.168.1.10:9000

If the coordinator is down at boot:

[pm] coordinator unreachable at http://192.168.1.10:9000 — will retry

Either way the Pi starts normally. The coordinator is purely additive — removing it changes nothing about how the Pi operates.

manual registration (curl)

If you're not using coordinator_url, you can register a node manually:

curl -X POST http://192.168.1.10:9000/register \
     -H "Content-Type: application/json" \
     -d '{"name":"pi-kitchen","host":"192.168.1.42","port":5001}'

---

log files

Each script writes stdout and stderr to a .log file in log_dir. For a script with key CAN-M the log is written to <log_dir>/CAN-M.log. Logs persist across restarts and are also tailable directly:

tail -f /var/log/jobs/CAN-M.log

The last 500 lines are buffered in memory and replayed instantly when you open the log panel in the browser. New lines stream in live via WebSocket.

---

API reference

Each Pi exposes the following endpoints:

method	path	description
GET	/	local dashboard UI
GET	/api/status	JSON status of all scripts
POST	/api/start/<key>	start a script
POST	/api/stop/<key>	stop a script
GET	/api/log/<key>	last 500 lines of log (plain text)
WS	/ws	status push every 2s + start/stop commands
WS	/ws/<key>	live log stream for one script

The coordinator exposes:

method	path	description
GET	/	fleet dashboard UI
POST	/register	register or re-register a node
GET	/api/fleet	full fleet status JSON
GET	/api/nodes	all node info
POST	/api/<node>/start/<key>	start a script on a specific node
POST	/api/<node>/stop/<key>	stop a script on a specific node
WS	/ws	live fleet status push + forwarded commands

---

architecture

[ browser ]
    │
    │  WebSocket  (live status + logs)
    │
    ├─────────────────────────┐
    │                         │
[ coordinator :9000 ]   [ Pi dashboard :5001 ]
    │                         │
    │  HTTP poll /api/status   │  asyncio subprocess
    │  every 3s                │  stdout/stderr → WS
    │                         │
    └──────── Pi nodes ────────┘
         pi-kitchen  :5001
         pi-garage   :5001
         pi-shed     :5001

Each Pi is standalone. The coordinator is optional. If the coordinator restarts, Pis re-register automatically within 30 seconds.

---

notes

• Scripts must run in the foreground. Processes that daemonize (fork and detach) will not be tracked correctly.
• State is in-memory. If the Pi's process_manager process restarts, running scripts become orphans. Use systemd to keep process_manager itself alive if uptime matters.
• sys.executable is recommended over "python" or "python3" to ensure the correct interpreter is used, especially inside a virtualenv.