Files
ducoterra d0077e0db4
Build and Push Containers / test (push) Successful in 45s
Build and Push Containers / build-and-push-app (push) Successful in 13s
phase: 09_wall_remote_page
**Phase 09 — final verification pass (all 3 tasks already in `complete/`; phase work was in the working tree, verified, nothing needed fixing)**

**Gates** — `uv run ruff check . && uv run ruff format --check .` → clean; `uv run pyright` → 0 errors; `uv run pytest` → **360 passed, 5 skipped** (only the realdisplay rung), coverage 96% total / 95% app.py; `MIRROR_REALDISPLAY=1 uv run pytest -m realdisplay` → 4 passed, 1 skipped (documented host limitation: no audio source on the real X11 path); `uv run pytest tests/e2e/test_remote_control.py` → 8/8 passed; `node --check` on every `static/*.js` → all pass.

**Completion criteria** — all met: (1) "Show on wall" stages on the display, asserted on display DOM + `/api/kiosk` ✓; (2) Back/Fullscreen + untouched second controller converge (TestConvergence) ✓; (3) `/api/streams` reports `viewers: 0` / `watching: 0` with only `/remote` open (TestACostsNothing) ✓; (4) followers count displays only — 0/1/1 with a second controller (TestFollowers) ✓; (5) controller works with no display: `followers: 0`, state stored (protocol test + e2e) ✓; (6) e2e + full `-m e2e` + realdisplay all green ✓; (7) README + SECURITY.md updated, Silverblue default now `/kiosk` (installer, conf, gnome script, checker) ✓; (8) suite 100% green, coverage >90% on new code ✓; (9) PLAN.md untouched; AGENTS.md changed only for a stale statement ✓.

**Documentation** — README.md reviewed in full against code (page table, quickstart, env-var table vs `config.py`/`debug.py`, presets, keys, deployment, kiosk section): no inconsistencies found. AGENTS.md: fixed the one stale statement introduced by this phase ("Four pages" → "Five pages"); no other staleness, nothing added.

**Decisions** — no repo changes were required; the one realdisplay skip is a pre-documented environment fact, not a failure.

**Next pending phase** — none; `todo/` held only phase 09.
2026-11-03 23:50:43 -05:00

286 lines
16 KiB
Markdown
Raw Permalink Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
# mirror
**Device-agnostic screen mirroring that lives entirely in the browser.**
`mirror` runs a small HTTP server on your machine. Open it from any device on
your network:
| Page | URL | What it does |
| ------------ | ------------------ | ----------------------------------------------------------------------- |
| **Share** | `/mirror` (also `/share`) | Picks a window, screen, or tab via the OS-native picker and mirrors it. |
| **Watch** | `/watch` | Shows every screen currently being mirrored; click a tile to take the stage. |
| **Kiosk** | `/kiosk` | A chrome-free wall for an unattended display: no header, no counts — the net pill is the only chrome, and the shared layout state (below) arranges it. |
| **Remote** | `/remote` | A controller for the shared wall, from a laptop or a phone: lists the live mirrors and the wall's state and sends the decisions (show this, back to the wall, full screen) — no video, no bandwidth. |
| **Index** | `/` | Signpost + live count of mirrors. |
Any number of people can mirror at the same time and any number of people can
watch. There are **no accounts and no passwords** — this is built for a trusted
local network, and that assumption is deliberate and documented in
[SECURITY.md](SECURITY.md). What *does* decide who gets in is the source address:
by default any address is welcome, and `--allow-cidr` / `MIRROR_ALLOW_CIDRS`
narrows that to the networks you name.
## Quickstart
```bash
uv sync --extra dev
uv run mirror serve # http://0.0.0.0:8055
```
Then, from another machine on the LAN:
```
http://<this-machine>:8055/mirror # to share
http://<this-machine>:8055/watch # to watch
http://<this-machine>:8055/kiosk # the unattended wall (kiosk display)
http://<this-machine>:8055/remote # drive the wall from this device
```
Linux is the primary target. On Linux the native picker is X11/Wayland's own
(Chromium uses `xdg-desktop-portal` on Wayland/GNOME/KDE, and its built-in X11
picker on Xorg), so nothing has to be installed on the client side.
The share page works on a phone as well as a desktop: `/mirror` is usable
down to a 320 px-wide viewport, and the preview keeps its shape but never
shrinks below the controls it holds, so the Share button is always on screen.
The target *client* setup for an always-on display is Fedora Silverblue running
GNOME Kiosk — see [`docs/kiosk-silverblue.md`](docs/kiosk-silverblue.md).
## How it works
Media never touches the server. Each mirror is a WebRTC peer connection
established directly between the sharer's browser and each watcher's browser
(a "mesh" fan-out). The server is a **signaling relay and roster only**: it
tracks who is publishing what, forwards SDP/ICE between peers, and pushes
roster deltas to viewers.
```
sharer browser ──ws──► ┌──────────────────────┐ ◄──ws── watcher browser
│ │ FastAPI │ ▲
│ offer/ICE ───►│ /ws (signalling) │◄─── offer/ICE │
│ │ registry (in-mem) │ │
└────── WebRTC (SRTP, LAN, p2p) ─────────┴─────────────────┘
```
**The wall has one shared state.** The server keeps which mirror is on stage, whether
it is fullscreen, and how many displays are obeying that (`followers` — peers whose
`hello` said `"kiosk": true`). Any connected peer may change it with an `arrange`
frame on `/ws` (`{"type":"arrange","staged":"<id>","fullscreen":true}`); an absent
field is "no opinion", `"staged": null` is "back to the wall". A change is pushed
immediately as a `layout` frame to every viewer (and back to the sender), and the
same object is readable at `GET /api/kiosk` — `curl`-debuggable like `/api/streams`,
`Cache-Control: no-store`. The state has no audio: sound always traces back to a
gesture on the device that plays it (see [SECURITY.md](SECURITY.md) §4). The
`/kiosk` page is the display that follows this state: a chrome-free wall that
subscribes to every mirror like `/watch` and does what the state says —
`Cache-Control: no-store` on the read and on the page itself. If a keyboard is
plugged in, the same keys as `/watch` work, and every keypress publishes the
same `arrange` intent, so a controller and a display converge on one state
(`m` is local-only: a display never becomes audible because another machine
asked).
### Driving an unattended wall
The shared state exists for a box with no keyboard: point the box at `/kiosk`,
open `/remote` on a phone, and press **Show on wall** — the box stages that
mirror, and **Back to the wall** and **Full screen** do the same for their
decisions. When a keyboard is plugged into the box, the wall keys still work
there, and a keypress performs the change *and* publishes the same `arrange`
intent, so a key on the box is seen by every controller — a controller and a
display converge on one state. Fullscreen is **requested, not promised**: a
remote command carries no user activation, so the browser may refuse the
request; the in-page stage fills the viewport anyway (and the kiosk box's
window manager forces the window fullscreen). The state lives in the server's
memory only: a restart returns the wall to a wall.
## Configuration
Environment variables (all optional; a `.env` file is loaded if present):
| Variable | Default | Meaning |
| ---------------------- | ------- | ---------------------------------------------------- |
| `MIRROR_HOST` | `0.0.0.0` | Bind address for `mirror serve`. |
| `MIRROR_PORT` | `8055` | Bind port. |
| `MIRROR_ICE_SERVERS` | `[]` | JSON array of RTCIceServer objects, if peers cannot reach each other directly. |
| `MIRROR_MAX_MIRRORS` | `0` | Cap on concurrent mirrors (`0` = unlimited). |
| `MIRROR_MAX_BITRATE` | `25000000` | Per-leg WebRTC bitrate cap in bits/s — the mesh uploads N × bitrate per sharer, so this is the ceiling. 25 Mbit/s suits a wired LAN at 1440p; a Wi-Fi sharer with many viewers may need less, not more. |
| `MIRROR_PING_INTERVAL` | `20` | Seconds between socket liveness pings. |
| `MIRROR_ALLOW_CIDRS` | *(unset)* | Comma-separated source networks allowed to connect, e.g. `10.2.0.0/22,fd00::/64`. Unset means unrestricted. A wall needs this range too, and its address is DHCP — give it a reservation. |
| `MIRROR_TRUSTED_PROXIES` | `127.0.0.0/8, ::1/128` | Which peers may speak for a client via `X-Forwarded-For`. Only a proxy that overwrites the header belongs here. |
| `DEBUGPY` | `0` | `1` opens a debugpy listener on `DEBUGPY_PORT`. |
| `DEBUGPY_PORT` | `5678` | debugpy port. |
| `DEBUGPY_WAIT` | `0` | `1` blocks the server at start until a debugger attaches. |
## Platform notes
**Screen capture needs a secure context.** `http://localhost` is one; a plain
`http://192.168.x.x` opened from another device is not, and the browser hides
`getDisplayMedia()` completely. The `/mirror` page detects this and prints the
three ways forward (share from localhost, serve HTTPS, or mark that one origin as
secure on the sharing device). Watching never needs permission.
This is why the fleet deployment terminates TLS in nginx rather than exposing
`mirror serve` directly — over `https://`, every device on the network can share
and none of this applies. A bare `mirror serve` is for the machine it runs on and
for testing.
**Linux.** On X11/Xorg Chromium uses its native capturer — the picker appears and
that is the end of it. On Wayland (GNOME/KDE) Chromium goes through
`xdg-desktop-portal`, so the compositor shows its **own permission dialog for
each share**, which mirror neither controls nor tries to bypass. Chrome's picker
offers whole screens and individual windows; on Wayland what the portal will
agree to depends on the compositor.
**Audio.** A share carries audio when the browser's picker offers it (Chrome's
"share audio" tickbox, or a tab's own audio); the sharer's panel says
`shared` / `silent` / `not offered`, and the wall shows an audio button only
when one exists. On the wall, only the staged mirror can be audible — **press
`m` on the wall to hear the staged mirror** (unmuting it mutes the others,
leaving the stage re-mutes it). What *system* audio a browser can capture is
a browser/platform fact, not a mirror one: Chrome on Linux gets it where
PipeWire offers it, Firefox's desktop capture does not offer it at all. See
[SECURITY.md](SECURITY.md) §4 "Audio".
**Quality.** The share page has a Quality control — `Sharp` (default),
`Balanced`, `Light` — because a sharer's machine is a resource the share
spends, and `MIRROR_MAX_BITRATE` alone cannot express "fewer pixels". Each
preset is one request to the capture device and one per-viewer encoder
budget:
| Preset | Requested pixel cap | Frame rate | Per-viewer bitrate | `degradationPreference` (what the encoder may sacrifice) |
| ------------ | ------------------- | ------------ | ------------------------------ | ------------------------------------------------------- |
| `Sharp` | ≤ 3840×2160 | 30 fps (up to 60) | 100% of `MIRROR_MAX_BITRATE` | `maintain-resolution` — frame rate |
| `Balanced` | ≤ 1920×1200 | 24 fps (up to 60) | 60% of `MIRROR_MAX_BITRATE` | `balanced` |
| `Light` | ≤ 960×600 | 15 fps (up to 30) | 35% of `MIRROR_MAX_BITRATE` | `maintain-framerate` — resolution |
`Sharp` is the default and is what the sharp-capture work shipped: its
request, cap and preference are the old hardcoded values verbatim, so nothing
it proved gets worse — the new presets only ever ask for less, and a preset
can never raise a leg above the server's ceiling. `MIRROR_MAX_BITRATE` stays
the ceiling the factor is applied to: a Wi-Fi sharer picks `Light` in the
page, an operator lowers the ceiling for everyone.
Who it is for: the software-encoding sharer. A Firefox/Linux box often has
no hardware video encoder (VA-API exposes no H.264 encode entrypoint on many
machines), where 4K at 30 fps burns roughly a whole core — `Light` trades
pixels and frame rate for a fan that does not spin, and under pressure the
encoder will in fact sacrifice what its `degradationPreference` allows.
Changing the preset applies to the share that is already running — the
per-viewer bitrate and frame rate always, and the capture size where the
browser allows a live rescale; when it does not, the panel says the size
lands on the next share instead of pretending it did. `q` cycles the preset.
The panel also shows the quality budget in force and the codec the legs
actually negotiated — `—` where a browser does not report it, never a guess.
## Deployment
The internal deployment is `https://mirror.reeseapps.com`, described and automated
in the deployments repo under `mirror/` (`app/` quadlet, `nginx/` edge, `certbot/`,
`route53/`, and a README with the VM SOP). Three things about it are worth knowing
from this side:
- **TLS ends at nginx on the service VM**, not in this process, so there is no
`--https` flag here. That is what makes `getDisplayMedia()` available on every
device, and it means an `Upgrade` header and a long idle timeout on `/ws` are
load-bearing.
- **The address list is enforced twice** — nginx `allow`/`deny`, and
`MIRROR_ALLOW_CIDRS` in the container for the path that bypasses nginx. Change
both in one commit.
- **The container has no internet egress.** Media goes browser-to-browser and the
server only relays JSON, so it dials nothing; `Internal=true` is what makes
that a fact rather than an intention.
### Kiosk client
A wall display is a Fedora Silverblue box running the locked-down GNOME Kiosk
session, auto-logged in as `kiosk`, with Firefox in `--kiosk` mode pointed at
the wall. It is set up and repaired entirely over SSH — no keyboard, no console,
no monitor. The display page is `/kiosk`, and it is the installer default
too; an existing `kiosk.conf` URL wins over the default, so moving a box is
the one-line re-point in [the runbook](docs/kiosk-silverblue.md#two-endpoints-watch-and-kiosk). A brand-new box layers the kiosk packages once first
(`rpm-ostree install gnome-kiosk gnome-kiosk-script-session`, then a reboot);
after that these two commands are both the setup and the repair path:
```bash
scp -r docs/kiosk-silverblue client:/var/tmp/ && ssh client \
'cd /var/tmp/kiosk-silverblue && sudo sh install-kiosk.sh && sudo systemctl restart gdm'
ssh client 'sudo sh /usr/local/share/mirror-kiosk/check-kiosk.sh'
```
Because nobody is standing at the machine, every claim about it is verified from
the outside: `loginctl` for the session, the foreground VT for "is the wall
actually on screen", the server's `/api/streams` for the viewer, the sharer's own
peer state for the media, and a probe page for the rendered resolution. The
runbook records why each file lives where it does on an
otherwise read-only OS.
**When it hangs.** A wall is unattended, so a hang has to end without a person:
the kernel turns lockups and oopses into a reboot, and PID 1 must keep kicking the
hardware watchdog or the chipset resets the box. Being *quiet* is a different
failure than being hung, so the server watches `/healthz` for that and pages over
ntfy ([`docs/wall-watch/`](docs/wall-watch/)). Both are installed by the commands
above and asserted by `check-kiosk.sh`.
**Keys.** The whole wall works from a keyboard — and the box needs no keyboard
attached to be operated *or* debugged; the keys are there when one is plugged
in. On `/watch`:
| Key | Action |
| --- | --- |
| `j` / `↓` · `k` / `↑` | Select the next / previous mirror (wraps). |
| `Enter` / `f` | Stage the selected mirror fullscreen (`f` re-enters fullscreen when already staged). |
| `g` | Back to the wall. |
| `n` / `p` | Stage the next / previous mirror. |
| `1`–`9` | Stage the Nth mirror. |
| `m` | Mute / unmute the staged mirror — on the selected mirror it stages it first. |
| `?` | Show / hide this list on screen. |
| `Esc` | Close the key list, then leave the stage. |
The same wall keys work on `/kiosk` too — the unattended display — and a
change made there (a keypress stages the mirror *and* publishes the same
`arrange` intent) is what every other display follows.
A staged mirror's caption bar and control cluster are **hidden at rest**, so
the shared video — a terminal's bottom line included — is never covered by
them. A pointer move over the stage (a tap, for touch) or any key press
while staged reveals them briefly; after ~3 s without activity they fade
back, so a parked pointer or a one-time key press never pins them up.
A mirror's audio is only heard while it is on the stage: every mirror starts
muted, unmuting one (the tile's audio button or `m`) mutes the others, and
leaving the stage re-mutes it. Sound never starts without a click or a
keypress.
On `/mirror`: `s` starts a share, `x` stops one, `q` cycles the Quality
preset, and `Esc` stops while live. Keys
with `Ctrl`/`Cmd`/`Alt` are never intercepted, and a key typed in the display-
name field is text, not a command. `?` is the only key that needs Shift, and it
is matched on the character itself — on a layout where `?` is a dead key the
list is not reachable, so keep this table to hand.
## Development
```bash
uv run pytest # unit + API + browser E2E (~95s)
uv run pytest -m "not e2e" # everything except the browser (~0.5s)
uv run ruff check . && uv run ruff format --check . && uv run pyright
```
The browser suite drives **real capture**: Chromium's fake media device yields a
moving monitor stream (1280×720 by default; clamped to the capturing window when
a size is requested, as `/mirror` now asks), so the assertions are on decoded,
changing pixels at the watcher — not on log lines.
There is a second, opt-in rung that exercises Chromium's **real X11 capture
backend** on a private Xvfb display, isolated from your own session so it can
never prompt you or mirror your desktop:
```bash
MIRROR_REALDISPLAY=1 uv run pytest -m realdisplay
```
See [AGENTS.md](AGENTS.md) for the operating rules,
[.agents/PLAN.md](.agents/PLAN.md) for the locked architectural decisions, and
[SECURITY.md](SECURITY.md) for why there is no login.