SnapshotView fetches one frame at a time through the backend, pauses
while the tab is hidden and backs off after failures. PrinterCard picks
it when /stream answers kind "snapshot".
The console could watch and control printers but not give them work. The
Jobs page uploads a .3mf/.gcode: it creates the job (POST /jobs), PUTs the
file straight to the presigned storage URL with a progress bar, and lists
the farm's jobs, polling every 5s since there is no job push channel yet.
The scheduler then starts it on the next idle printer, so there is no Start
button - a queued job IS the request.
Deliberate details: the PUT carries no Authorization header (the URL is
presigned and an uncovered header makes storage reject it) and the same
Content-Type the URL was signed for; a failed upload deletes the job row
it just created, so no queued job waits for a file that never arrives;
only queued/uploading jobs offer Delete, matching what the backend allows.
JobUpload and JobList live in @continuum/ui and take the API calls as
props, like the printer card. A shared apiErrorMessage() replaces the
inline copy on the fleet page.
Checked in real Chrome against the live stack with job dispatch OFF: upload
to the real bucket (57 bytes landed intact), listed as queued, still queued
after a scheduler tick, Delete clears it, no page errors; the test object
was removed from the bucket afterwards. pnpm -r typecheck is clean.
The player posted its SDP offer to the gateway's go2rtc directly, which
cannot work from a browser on another origin: no CORS headers on any
go2rtc response, a preflight required because application/sdp is not
CORS-safelisted, and the password checked before CORS so that preflight
gets a 401. It now hands the offer to a `signal` function the app
supplies, which relays it through the backend.
Pinned to iceTransportPolicy "relay" whenever ICE servers are present,
so media always goes through the TURN relay and never peer-to-peer. A
hosted console cannot rely on reaching a customer's gateway directly,
and a camera that works from the workshop but not from home is worse
than one that behaves the same everywhere.
With no TURN configured it falls back to "all", which still works for a
viewer on the gateway's own network. Relay-only with no relay would
gather no candidates and fail every time, so the fallback is what makes
an unconfigured deployment degrade rather than break - and a failed
connection in that state now says so, since the browser reports nothing
but "failed" and the fix is a deployment setting.
It also waits for ICE gathering before signalling rather than trickling:
the answer comes back through a single request/response relay hop, so
there is no channel to trickle over afterwards and the offer has to
carry its candidates. Bounded at 3s, because gathering can hang on a
candidate that never resolves and signalling with what we have beats
waiting forever.
`signal` is a prop rather than a fetch in the component, like `stream`
already was: @continuum/ui stays plain Vue that knows nothing of the API,
which is what lets it render from a test or the desktop shell.
Auth is Supabase end to end and always was: the backend verifies access
tokens asymmetrically against the project's published JWKS and holds no
signing key of its own, and `requireUser` wants that token in
`Authorization: Bearer`. Nothing about that needed changing.
What was wrong is only here. The two REST calls forwarded
`useRequestHeaders(["cookie"])`, which is what you do when a Nuxt server
is proxying a browser's request - and this app sets `ssr: false` so the
same static bundle can serve the browser and the Tauri shell. There is no
server to forward anything, so the header arrived empty and every call
should have been getting a flat 401.
The token was to hand the whole time. `stores/telemetry.ts` has been
reading `session.access_token` for its socket since it was written; the
REST side simply never did.
`useApiFetch` is now the one way this app calls the backend, and it reads
the token inside the fetcher rather than capturing it at setup, because
supabase-js refreshes the session in the background about once an hour - a
captured token would work until it quietly stopped. With no session it
omits the header entirely rather than sending an empty Bearer, which the
backend rejects with a message about the header's shape and reads like a
client bug; getting the user a session is the Supabase module's redirect's
job.
The socket keeps using a `?token=` query parameter, which is not an
inconsistency: a browser WebSocket cannot set request headers.
`nuxt typecheck` clean.
`WebRTCStreamView` has existed since before this commit with no consumer:
nothing rendered it, and the URL it would have used came from a
`go2rtcStreamBase` in `nuxt.config.ts` defaulting to
`http://localhost:1984`.
That default was the real problem. A camera stream lives on a gateway box
inside one farm's network, each farm has its own, and the gateway now
reports its address to the backend when it connects. A compiled-in
constant could only ever be correct for a single-farm install, and its
localhost default pointed every player at the operator's own laptop. It
is gone, replaced by `fetchPrinterStream`, which asks
`GET /printers/:id/stream`.
Absence is the normal case here, not an error: only Bambu's X1/H2 class
publishes a camera URL a gateway can restream, so a P1 or A1 has a camera
and nothing to connect to. The backend 404s for those, this resolves to
null, and the card renders no player - offering one that can never load
would be worse than offering nothing. The lookup also runs only for
printers whose `cameraStream` the fleet list already reports, so a
twenty-printer farm does not make twenty requests to learn that eighteen
of them 404.
`PrinterCard` takes the stream as a prop rather than fetching it, like
`live` above it: `packages/ui` is plain Vue that knows nothing about the
API or about Nuxt, and that is what lets these components render from a
test or the desktop shell with no server behind them.
Three things in `WebRTCStreamView` that would have bitten whoever first
pointed it at a real printer:
- The fetch status was never checked. go2rtc answers a request for a
stream it does not have with an error page, and handing that to
`setRemoteDescription` throws about SDP parsing - which sends you
looking at the wrong layer entirely.
- `connect()` is async inside a watcher with nothing catching it, so a
failure left the badge on "connecting" forever and showed up only as an
unhandled rejection in the console.
- Two prop changes in quick succession raced: the slower request could
finish last and apply its answer to a connection already replaced and
closed, leaving a black video with a "connected" badge.
No `iceServers`, deliberately and now commented: both ends are on the
same LAN, so host candidates are enough. The same fact is the MVP's
limitation - there is no relay, so this plays for someone on the
printers' network and for nobody else.
`nuxt typecheck` and `vue-tsc --noEmit` both clean.
octoturge