I Replaced My Monitoring Dashboard With a Factory Warning Light

#networking #monitoring #observability #hardware

I replaced my monitoring dashboard with a factory warning light.

Not metaphorically. Literally.

It's a Patlite LA6-POE — the kind of stacked warning light you see on Japanese factory floors that flashes red when the line stops. Five segments, one PoE cable, ~30 cm tall.

Mine doesn't warn. It shows.

The Inversion

Warning lights are designed to stay dark until something is wrong, then scream.

So I inverted it.

I wanted a light that stays always on, showing current bandwidth as a stack. The more layers lit, the more the network is active.

When 1 segment is lit, things are quiet.

When all 5 are lit and the top one flashes, someone is pushing a lot of traffic.

I barely open the dashboard anymore. The corner of the room tells me what I need to know.

The Build

Hardware: Patlite LA6-POE, one PoE switch port (power + data in a single cable)
Software: ~200 lines of Python on the monitoring host
Protocol: Patlite's obscure PNS binary protocol over TCP/10000. Reverse-engineered from the manufacturer PDFs. One round-trip is ~11 ms.

Data path:

PRTG (sFlow/SNMP from edge router)
    ↓ XML API polled every 5 seconds
    ↓ Python daemon + hysteresis logic
    ↓ PNS command over TCP
    ↓ Patlite LA6-POE

Colors (fixed by hardware):

Bottom → Blue → Green → Yellow → Pink → Red (top)

The PRTG Trap

PRTG's XML for traffic sensors contains <lastvalue_raw> twice per channel:

First: cumulative bytes since sensor start
Second: current rate in bytes/second

# WRONG — you will see 8 million Mbps and question reality
rate = channel.find('lastvalue_raw').text

# RIGHT
raw_values = channel.findall('lastvalue_raw')
rate_bps = float(raw_values[1].text)

I lost half a day.

This kind of bug is the worst — it doesn't crash, it just lies.

Thresholds with Hysteresis

I didn't design this for precision. I designed it for perception.

Naive thresholds make the light flicker like a broken Christmas tree. Hysteresis (separate UP/DOWN thresholds with ~10% gap) fixes it.

Level	UP (Mbps)	DOWN (Mbps)	Segments	Flash	Buzzer
0	—	< 18	none	—	—
1	≥ 20	< 27	1	—	—
2	≥ 30	< 45	2	—	—
3	≥ 50	< 72	3	—	—
4	≥ 80	< 90	4	—	—
5	≥ 100	< 135	5	ON	—
6	≥ 150	< 135	5	ON	0.6s chirp

Result

This device solves zero tickets.

Sends zero notifications.

Improves no SLAs.

But I open the dashboard far less — and I notice problems earlier.

It does nothing useful, and everything I wanted.

It turned a factory warning light — originally designed to scream until somebody makes the screaming stop — into a device I actually want to glance at.

And that, in 2026, is more than enough.

(Six months ago, I solved a Cloudflare Wrangler problem with X11 forwarding instead of API tokens. Apparently this is my brand now: solving modern problems with historically inappropriate tools. This is the second time.)

If people are interested, I can share the exact setup and code.