Foudra is a Manufacturing Resource Planning (MRP II) system for managing the shop floor. It schedules purchase orders and work orders to build what you want, when you want it. It has an extensive API for adding third-party integrations, like this.
A daemon that drives 3D printers — Prusa printers over PrusaLink, and anything behind OctoPrint — and keeps the Foudra MRP honest about what they made and what they ate. Runs on a Raspberry Pi beside the printers. Well, it runs on whatever hardware you have lying around that can access the Foudra API over the Internet, and your printer(s).
One daemon drives the whole farm, each printer in a goroutine of its own, so a slow upload to one never holds up the others.
First you'll need an API key on Foudra. Get one here: https://yourcompany.foudra.com/api-keys.
A Pi Zero 2 W on the 64-bit Raspberry Pi OS is plenty: the binary is
a few MB and idles at a few MB of memory. (make build-pi32 for the
32-bit image.)
make build-pi && scp bin/pi/print-manager pi:/usr/local/bin/
ssh pi 'sudo useradd -r -s /usr/sbin/nologin print-manager; \
sudo mkdir -p /etc/print-manager /var/lib/print-manager/gcode; \
sudo chown -R print-manager /var/lib/print-manager'
scp deploy/print-manager.service pi:/tmp/ && ssh pi 'sudo mv /tmp/print-manager.service /etc/systemd/system/'
# /etc/print-manager/env (root:root 0600):
# MRP_URL=https://yourcompany.foudra.com
# MRP_API_KEY=<Your new API key - something like fdr_xxxxx>.
# PRINTER_KEY_CORE_ONEP_1=<The private key for your printer>.
ssh pi 'sudo systemctl daemon-reload && sudo systemctl enable --now print-manager && journalctl -fu print-manager'Copy sliced files and their sidecars into /var/lib/print-manager/gcode/.
make build # for this machine, into ./bin
make build-pi # linux/arm64 static binary, into ./bin/pi
make check # fmt, vet, testCopy bin/pi/print-manager to the Pi with the environment file that
deploy/print-manager.service expects.
The daemon needs two things from the environment and discovers the rest:
| Variable | Meaning |
|---|---|
MRP_URL |
https://yourcompany.foudra.com |
MRP_API_KEY |
A key made under the MRP's API keys page, named for the printer farm |
PRINTER_KEY_<CODE> |
One per printer, named by its workstation code upper-cased, other characters as _: PRINTER_KEY_CORE_ONEP_1. |
GCODE_DIR |
Where a part's G-code name resolves; default /var/lib/print-manager/gcode |
POLL_INTERVAL |
Default 15s |
Which printers exist, and where, comes from the MRP: every active,
automatic workstation whose Custom data names a printer url is
driven. That JSON also says which driver to use, and carries the slicer
profile and the per-printer policy. A Prusa:
{"driver": "prusalink", "url": "http://192.168.1.102", "profile": "core-one-0.4",
"window": {"start": "08:00", "end": "22:00", "timezone": "Europe/Dublin", "must_finish_in_window": true},
"filament_policy": {"change_for_priority_at_or_above": 1}}A printer behind OctoPrint:
{"driver": "octoprint", "url": "http://octopi.local", "profile": "mk3s-0.4",
"window": {"start": "08:00", "end": "22:00", "timezone": "Europe/Dublin", "must_finish_in_window": true}}driver defaults to prusalink, and prusalink_url is still read in
place of url, so stations set up before there was a choice need no
change. profile is any name you like: it picks the G-code from the
part (below), so use one per printer type and nozzle. The window and
filament_policy blocks are optional.
The daemon also writes its own bookkeeping into this JSON
(awaiting_clear_job, last_completed_job, run). Leave those alone
while it is running.
-
The key: in OctoPrint, Settings → Application Keys (or your user's API key under User Settings). Put it on the Pi as
PRINTER_KEY_<CODE>. -
G-code, not bgcode: OctoPrint printers take plain
.gcode, so give the part a G-code under the OctoPrint printer's profile:{"gcode": {"mk3s-0.4": "bracket_PETG_MK3S.gcode"}}. -
Uploaded files are replaced. Before each upload the daemon deletes any file of the same name in OctoPrint's local storage, then uploads afresh. The deletion clears the file's print history, which is how the daemon tells "a print has finished and its parts are on the bed" from "ready for a job" — OctoPrint itself just says Operational for both. Don't keep files you care about under the names the daemon uses.
-
No printer connected (USB unplugged, printer off) reads as needing attention: the daemon logs it and starts nothing.
-
Testing without a printer: OctoPrint's bundled virtual printer is enough. With an OctoPrint running and connected to it,
export OCTOPRINT_URL=http://127.0.0.1:5000 OCTOPRINT_KEY=… go test -p 1 -run TestLive -v ./internal/octoprint/ ./internal/farm/
prints, finishes, clears and cancels on it for real, and runs the farm loop through two beds.
-p 1because both share the one printer. Without those variables the live tests are skipped.
A printed part's G-code is named in its custom data, by profile:
{"gcode": {"core-one-0.4": "bow_node-case_PP_COREONE.bgcode"}} — a bare
name resolves under GCODE_DIR, a URL is fetched there once. The
sidecar .json that scripts/slice.sh writes beside it supplies
per_bed, est_seconds (for the window) and filament_g (the actual).
slice.sh -n 4 part.stl puts four copies on the bed (PrusaSlicer
arranges them), names the file …_x4.bgcode and writes per_bed: 4:
each print then completes four units and charges the whole bed's
filament to them. The render is still one part; the layout the slicer
chose is not known to Blender.
A printer key never goes in the MRP — it is full control of a hot
machine — so it lives only in /etc/print-manager/env on the Pi.
print-manager printers shows what would be driven and why anything
was skipped.