A start_print is acknowledged as soon as the gateway accepts it, and the
download and transfer that follow can still fail. That late result was only
logged, so the job sat in uploading until the stall reaper gave up twenty
minutes later.
The dispatcher now remembers which job each start command belonged to. A
late failure for it fails the job with the gateway's own reason in
metadata.lastDispatchError - but only while the job is still uploading, so a
job the printer has since started is left to its telemetry.
scripts/e2e-vendors.sh gains a phase where the proxy refuses the download
and the job must fail at once with that reason.
POST /farms/:id/printers now takes any of the four vendors the gateway can
drive. Credentials depend on the vendor: Bambu needs an access code and
serial, OctoPrint and PrusaLink an API key, Klipper nothing. A missing one
is a 422 naming what is needed. The API key is relayed to the gateway and
kept out of logs and errors like the access code.
The printer_vendor enum gains octoprint (migration 0006); PrusaLink is
stored as prusa, the name the enum already had.
scripts/e2e-vendors.sh adds each kind through the API against a fake REST
printer (scripts/fake-http-printer.ts) and runs a job through the real
scheduler to completion. The fake shares its author's reading of the docs
with the driver, so this proves the plumbing, not the protocols.
octoturge