CHG-0027: The Beacon on an ESP32


Date	2026-10-18
Change type	Upgrade
Status	Complete 2026-10-18
Systems	`turntable-power-01`: a new board, a plain ESP32, in place of the ESP32-S2
How	`make build flash provision` in `firmware/power-beacon`, by cable
Risk	Low. Most likely to go wrong: the old S2 powered at the same time as the new board. Both log in as `turntable-power-01`, the broker takes the account from one for the other, and each time says `offline` for the turntable
Decisions	ADR-0023, unchanged: any ESP32 with WiFi and 4 MB will do
Related	CHG-0015, the first beacon; CHG-0025, where the S2 needed an external antenna to join from the turntable

Summary

The beacon moves from an ESP32-S2 to a plain ESP32 that fits the case its owner bought. The case has no opening for an external antenna, and the S2 could not join from the turntable without one. The new board’s antenna is on the board.

Before anything was built for it, the board was flashed with ESP-IDF’s WiFi scan example, from a scratch copy, and read at the turntable, on a cable beside it: DVNTM-IOT at −58 to −59 dBm, DVNT-CORE at −45 to −48 dBm, over four scans. It stays on DVNTM-IOT and the mobile broker for now.

It is the same beacon, turntable-power-01: the same account, the same topics, the same ota grant. Nothing in the tenant changes.

The firmware’s target changes, and so does make chip. It used to stop unless the board was an S2, which is what kept it from writing over the cover screen. Every board EdS has is now a plain ESP32, so it checks the MAC instead.

Goal

make chip stops for any board but 70:4b:ca:27:06:14.
make flash writes the bootloader with rollback, the partition table and the app; make provision writes the creds partition.
At the turntable, on its outlet, in its case: system.boot and wifi.connected from turntable-power-01, with the RSSI, and the console showing it online. nowplayd logs it switched on.
Switched off with the turntable: offline within about fifteen seconds.

Scope

In scope: the new board, and the firmware’s target and make chip. Out of scope: moving the beacon to DVNT-CORE and the home Pi (PLAN-one-pi-backend.md); the stand and the screen.

Risk and impact

Risk	Where	Guard
The old S2 powered with the new board	the broker, nowplayd, the stand	The S2 off its outlet before the new board is provisioned; erased afterwards
The wrong board written	the Mac’s serial port	`make chip` checks the MAC
It does not join from the outlet in its case	the beacon	Read at the turntable already; its `wifi.connected` carries the RSSI. The S2 back on its outlet undoes it

Procedure

Step 1: Flash

Nothing in service changes: the board has never held anything of EdS’s.

Run:

cd firmware/power-beacon
. ~/.espressif/tools/activate_idf_v6.1.sh
make build flash

Verify:

make chip names the board as this beacon; the flash verifies.
The serial log: the app starts, unprovisioned.

Step 2: Provision

Needs the tenant’s Terraform outputs, so the Mac on the kit’s network. The S2 is unplugged first.

Run:

make provision

Verify:

make monitor: WiFi joined, the broker reached, online published.

Step 3: On its outlet

The board in its case, on the turntable’s outlet.

Verify:

The turntable switched on: wifi.connected with its RSSI; online on the console; nowplayd logs the source’s equipment switched on.
Switched off: offline, and the stand as ADR-0028 says.

Verification

The Goal’s checks, and a day on the outlet without the connection coming and going.

Undo

The S2 was erased after the change, so undo is a build for it and a flash and provision by cable: CONFIG_IDF_TARGET back to esp32s2, and CHIP=esp32s2 MAC=7c:df:a1:3c:d6:62 PORT=/dev/cu.usbserial-110 to make. Then it goes on the outlet, with its external antenna, and the new board off it.

Outcome

Completed after the change has run.

When	Step	What happened
2026-10-18	before	ESP-IDF’s scan example, repeating every three seconds, flashed from a scratch copy; nothing of it committed. The board had come with a factory image, no EdS partition table. At the Mac: DVNT-CORE −36 to −40 dBm, DVNTM-IOT −70. At the turntable, on a cable: DVNT-CORE −45 to −48, DVNTM-IOT −58 to −59.
2026-10-18	1	`make build flash` from `522e5fc`, on `/dev/cu.usbserial-0001` at the turntable: `make chip` named it this beacon; bootloader, partition table, `otadata` and app written and verified. `make chip MAC=00:11:22:33:44:55` refused it. It boots `522e5fc` from `ota_0`, says `not provisioned`, and keeps trying DVNTM-IOT with no key.
2026-10-18	2	First try not run: the Mac was on the home network, where the tenant’s state cannot be read (`tfstate.mobile.deevnet.net` does not resolve). Waits for the Mac on DVNTM-TD.
2026-10-18 13:25 UTC	2	The S2 unplugged; the Mac on DVNTM. `make chip`, a dry run, then `provision_tenant.py` with the main checkout’s tenant directory: the `creds` partition written. At the turntable, on the Mac’s cable: joined DVNTM-IOT on the first attempt, −65 dBm, as 10.20.30.201; `online` published. In the log store: `system.boot` (firmware `522e5fc`), `wifi.connected`, `mqtt.connected`. The console: `turntable-power-01` online.
2026-10-18 13:27 UTC	3	Taken off the Mac’s cable: nowplayd went idle with nothing switched on (`switched_on=false`, 13:27:32).
2026-10-18 13:29 UTC	3	In its case, on the turntable’s outlet, switched on: up 2.7 s after power (`mqtt.connected` at that uptime), first attempt, −66 dBm joining and −64 at the broker. nowplayd: `a source's equipment was switched source=turntable on=true` at 13:29:14.460, `current` stopped with the turntable named, the room waiting for a record. The console: online.
2026-10-18 13:30 UTC	3	Switched off at the outlet: the broker sent the will, and nowplayd logged `source=turntable on=false` at 13:30:51.7; `current` stopped with no source; the console `offline`. Fifteen seconds later (`NP_IDLE_AFTER`) nowplayd went idle with nothing switched on, and lightd sent the stand `look=""`, brightness 0, at 13:31:06.7: dark, from standby at 0.15.
2026-10-18	Verification	Every Goal check met but the day on the outlet, left as a follow-up.
2026-10-18	after	The old S2 on the Mac, `/dev/cu.usbserial-110`: ESP32-S2 revision v0.0, MAC `7c:df:a1:3c:d6:62`. Its whole flash erased with `esptool erase-flash`; the `creds` partition’s range read back all `0xFF`. It had not reached the broker before the erase: no event from it, nothing switched in nowplayd’s log. No backup of its factory image was ever taken, so erased is as near factory as it gets.

Departures from the plan

Follow-ups

A day on the outlet without the connection coming and going.
The old S2’s MAC recorded, and the board erased, credentials and all.