Prove video flows: an e2e suite with a real WebRTC viewer

e2e-edge.sh proves the camera signalling relay round-trips, and says so
honestly - it posts a stub SDP with go2rtc absent and accepts a 503 that
came back from the gateway. But "the offer round-trips" says nothing
about whether anyone would see a picture, and it once read as more than
it was: a stray go2rtc left over from an earlier run answered the stub
with an HTTP 500 and the test still printed ok. Video had never been
shown to flow.

scripts/e2e-camera.sh runs the real backend, the real proxy and the real
go2rtc against the H2C, and scripts/e2e-camera-peer.ts plays the
browser's part with werift: createOffer, relay through the backend,
setRemoteDescription, then ICE, DTLS and SRTP, counting RTP out the far
end. Measured: 124-149 packets and ~119-147 KB of payload within a
second or two. It sends no printer command; the camera is a pull and the
MQTT subscription is passive. It refuses to start if a go2rtc is already
running, and fails if one survives cleanup.

werift is a devDependency for that one file; nothing under src/ imports
it. Its 30 resolved packages were audited against LICENSES.md: 26 MIT,
Apache-2.0, BSD-3-Clause, Unlicense, and one MPL-2.0 (mediabunny), which
the policy allows while we do not edit its files.

Smaller fixes in the same pass:

- e2e-edge.sh now unsets the TURN variables. It asserts TURN is off, but
  sourcing dev-secrets.sh turns it on, and the failure message then
  printed the live Cloudflare username and credential. The new script
  prints only the camera id and a count, never the /stream body.
- the last hardcoded `psql -h localhost` in e2e-edge.sh now uses
  DATABASE_URL, like the rest.
This commit is contained in:
2026-10-23 16:01:47 +00:00
parent c312e37b4a
commit ac484a99ba
4 changed files with 365 additions and 3 deletions
+3 -2
View File
@@ -29,8 +29,9 @@
"ioredis": "^5.4.1" "ioredis": "^5.4.1"
}, },
"devDependencies": { "devDependencies": {
"drizzle-kit": "0.24.2",
"bun-types": "^1.1.10", "bun-types": "^1.1.10",
"typescript": "^5.6.2" "drizzle-kit": "0.24.2",
"typescript": "^5.6.2",
"werift": "^0.25.1"
} }
} }
+128
View File
@@ -0,0 +1,128 @@
/**
* A WebRTC receiver that is not a browser, for scripts/e2e-camera.sh.
*
* What it proves that nothing else here does: that video actually arrives.
* The signalling relay has always been testable with a canned SDP, and the
* go2rtc-to-printer leg was confirmed by hand, but "the offer round-trips"
* says nothing about whether a viewer would see a picture. This runs the
* browser's half of the exchange for real - createOffer, relay through the
* backend, setRemoteDescription, then ICE, DTLS and SRTP to go2rtc - and
* counts the RTP packets that come out the other end.
*
* It also exists because the suite's earlier stub offer (`v=0 ... t=0 0`, no
* media section, no ice-ufrag) could only ever be refused, and the test
* counted that refusal as success. A real offer has to be accepted.
*
* Results are printed as `ok<TAB>label` / `FAILED<TAB>label`, the convention
* run_probe in e2e-real-services.sh also reads. werift is a devDependency of
* this repo for this file alone; nothing under src/ imports it.
*
* Environment: BASE, PRINTER_UUID, ACCESS_TOKEN. The token never reaches a
* log line.
*/
import { RTCPeerConnection, RTCRtpCodecParameters } from "werift";
const base = process.env.BASE!;
const printer = process.env.PRINTER_UUID!;
const token = process.env.ACCESS_TOKEN!;
const MIN_PACKETS = Number(process.env.MIN_RTP_PACKETS ?? 50);
const WAIT_MS = Number(process.env.MEDIA_WAIT_MS ?? 25_000);
let failed = false;
const say = (good: boolean, label: string, detail = "") => {
if (!good) failed = true;
console.log(`${good ? "ok" : "FAILED"}\t${label}${detail ? ` (${detail})` : ""}`);
};
// werift offers only VP8 by default, and a Bambu camera is H.264 (High, level
// 4.1), so go2rtc refused the first version of this offer with "codecs not
// matched: video:H264 => video:VP8" - correctly. Every browser offers H.264,
// so offering it here is what a real viewer does, not a concession to the test.
// The parameters are Chrome's: packetization-mode=1, constrained baseline.
const pc = new RTCPeerConnection({
iceServers: [],
codecs: {
video: [
new RTCRtpCodecParameters({
mimeType: "video/H264",
clockRate: 90000,
rtcpFeedback: [{ type: "nack" }, { type: "nack", parameter: "pli" }, { type: "goog-remb" }],
parameters: "level-asymmetry-allowed=1;packetization-mode=1;profile-level-id=42e01f",
}),
],
},
});
const transceiver = pc.addTransceiver("video", { direction: "recvonly" });
let packets = 0;
let bytes = 0;
transceiver.onTrack.subscribe((track) => {
track.onReceiveRtp.subscribe((rtp) => {
packets += 1;
bytes += rtp.payload.length;
});
});
const offer = await pc.createOffer();
await pc.setLocalDescription(offer);
const offerSdp = pc.localDescription!.sdp;
say(/^m=video /m.test(offerSdp) && /^a=ice-ufrag:/m.test(offerSdp),
"the offer has a video section and an ice-ufrag, unlike the stub it replaces");
// A player that gets "stream not found" just asks again, and so does this: the
// gateway advertises a camera as soon as the printer reports its URL, a moment
// before go2rtc has been restarted with the stream in its config. Bounded, and
// the attempt count is reported, so a window that grows shows up as a number
// rather than as a flaky pass. Only that one reason is retried - a codec
// mismatch or any other refusal is a real failure and must fail at once.
const RETRY_MS = Number(process.env.OFFER_RETRY_MS ?? 20_000);
const started = Date.now();
let attempts = 0;
let res: Response;
let bodyText = "";
do {
attempts += 1;
res = await fetch(`${base}/printers/${printer}/webrtc`, {
method: "POST",
headers: { authorization: `Bearer ${token}`, "content-type": "application/json" },
body: JSON.stringify({ sdp: offerSdp }),
});
bodyText = await res.text();
if (res.status !== 503 || !bodyText.includes("stream not found")) break;
await Bun.sleep(1000);
} while (Date.now() - started < RETRY_MS);
say(res.status === 200, "the relayed offer is answered 200",
`HTTP ${res.status} after ${attempts} attempt(s), ${Math.round((Date.now() - started) / 1000)}s`);
if (res.status !== 200) {
// The gateway's reason is the whole diagnosis, and it carries no secrets.
console.log(`# ${bodyText.slice(0, 300).replace(/\s+/g, " ")}`);
await pc.close();
process.exit(1);
}
const answer = (JSON.parse(bodyText) as { sdp?: string }).sdp ?? "";
say(/^m=video /m.test(answer) && /H264\/90000/i.test(answer),
"the answer negotiates H264 video, which is what the printer sends");
await pc.setRemoteDescription({ type: "answer", sdp: answer });
const connected = await new Promise<boolean>((resolve) => {
const timer = setTimeout(() => resolve(false), WAIT_MS);
pc.connectionStateChange.subscribe((state) => {
if (state === "connected") { clearTimeout(timer); resolve(true); }
if (state === "failed") { clearTimeout(timer); resolve(false); }
});
if (pc.connectionState === "connected") { clearTimeout(timer); resolve(true); }
});
say(connected, "ICE and DTLS completed with go2rtc", `state=${pc.connectionState}`);
if (connected) {
const deadline = Date.now() + WAIT_MS;
while (packets < MIN_PACKETS && Date.now() < deadline) await Bun.sleep(250);
}
say(packets >= MIN_PACKETS,
`video RTP reached the viewer: at least ${MIN_PACKETS} packets`,
`${packets} packets, ${bytes} payload bytes`);
await pc.close();
process.exit(failed ? 1 : 0);
+229
View File
@@ -0,0 +1,229 @@
#!/usr/bin/env bash
# End-to-end test that a camera actually delivers video through the whole
# stack, not just that its signalling round-trips.
#
# e2e-edge.sh covers the relay with go2rtc deliberately absent: it posts a
# stub offer and treats a 503 that came back *from the gateway* as proof the
# message travelled. That is honest about what it proves, which is only the
# round trip. It cannot say a viewer would see anything, and it once read as
# more than it was because a stray go2rtc left over from an earlier run
# answered the stub with an HTTP 500 and the test still printed "ok".
#
# This one runs the real thing end to end:
#
# werift peer (stands in for the browser)
# -> POST /printers/:id/webrtc on the real backend
# -> /ws/edge/v1 -> the real continuum-proxy -> its real go2rtc
# -> RTSPS to the printer's camera
# and the RTP that comes back is counted.
#
# SAFETY: read-only against the printers. It sends no printer command of any
# kind; the proxy's MQTT subscription is passive and the camera is a pull of
# the existing stream. Printer credentials stay in continuum-proxy/printers.toml
# (gitignored) and are never read here beyond the id and name lines.
#
# Needs: a printer with an RTSP camera (an H2C - a P1P exposes none, see
# continuum-pending-hardware-tests finding 14), go2rtc, and these ports free:
# 1984 (go2rtc API, loopback) and 8555 (its WebRTC listener).
#
# Usage:
# scripts/dev-env.sh up && scripts/dev-env.sh schema
# scripts/e2e-camera.sh
#
# CAMERA_ID=h2c the gateway id (in printers.toml) with a camera
# GO2RTC_BIN=~/.local/bin/go2rtc
# PROXY_DIR=../continuum-proxy
# KEEP_DATA=1 leave the seeded farm behind for inspection
set -euo pipefail
REPO_ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
cd "$REPO_ROOT"
CAMERA_ID="${CAMERA_ID:-h2c}"
GO2RTC_BIN="${GO2RTC_BIN:-$HOME/.local/bin/go2rtc}"
PROXY_DIR="${PROXY_DIR:-$REPO_ROOT/../continuum-proxy}"
PORT="${PORT:-3040}"
AUTH_PORT="${AUTH_PORT:-3041}"
BASE="http://localhost:$PORT"
DATABASE_URL="${DATABASE_URL:-postgres://continuum:continuum@localhost:5432/continuum}"
VALKEY_URL="${VALKEY_URL:-redis://localhost:6379}"
EDGE_JWT_SECRET="${EDGE_JWT_SECRET:-e2e-edge-secret}"
# Nothing here presigns or uploads; the backend just refuses to boot without
# object-storage credentials (src/s3/index.ts).
R2_ACCESS_KEY_ID="${R2_ACCESS_KEY_ID:-e2e}"
R2_SECRET_ACCESS_KEY="${R2_SECRET_ACCESS_KEY:-e2e}"
R2_ENDPOINT="${R2_ENDPOINT:-http://localhost:9000}"
export DATABASE_URL VALKEY_URL EDGE_JWT_SECRET PORT
export R2_ACCESS_KEY_ID R2_SECRET_ACCESS_KEY R2_ENDPOINT
WORK="$(mktemp -d)"
BACKEND_PID=""
PROXY_PID=""
AUTH_PID=""
fail() { echo "FAIL: $*" >&2; exit 1; }
ok() { echo " ok: $*"; }
psql_q() { psql "$DATABASE_URL" -tAq -c "$1"; }
new_uuid() { cat /proc/sys/kernel/random/uuid; }
USER_ID="$(new_uuid)"
# The proxy supervises go2rtc, so it is the proxy's PID that has to die, and
# go2rtc with it. Anything that survives holds a printer access code in memory
# and answers the next run's offers - that exact leftover is how this script
# came to exist - so cleanup checks for it rather than assuming.
cleanup() {
local rc=$?
for pid in "$PROXY_PID" "$BACKEND_PID" "$AUTH_PID"; do
[ -n "$pid" ] && kill "$pid" 2>/dev/null || true
done
for _ in $(seq 1 20); do
[ -n "$PROXY_PID" ] && kill -0 "$PROXY_PID" 2>/dev/null || break
sleep 0.3
done
for pid in "$PROXY_PID" "$BACKEND_PID" "$AUTH_PID"; do
[ -n "$pid" ] && kill -9 "$pid" 2>/dev/null || true
done
if [ -z "${KEEP_DATA:-}" ] && [ -n "${ORG_ID:-}" ]; then
psql_q "DELETE FROM organisations WHERE id = '$ORG_ID'" >/dev/null 2>&1 || true
fi
if [ $rc -ne 0 ]; then
echo; echo "--- backend log (tail) ---"; tail -20 "$WORK/backend.log" 2>/dev/null || true
echo "--- proxy log (tail) ---"; tail -30 "$WORK/proxy.log" 2>/dev/null || true
fi
rm -rf "$WORK"
# A go2rtc that outlived us is a leak, not a detail.
sleep 0.5
if ps -eo comm | grep -qx go2rtc; then
echo "WARNING: a go2rtc process is still running after cleanup:" >&2
ps -eo pid,etime,cmd | awk '$3 ~ /go2rtc/' >&2
[ $rc -eq 0 ] && rc=1
fi
exit $rc
}
trap cleanup EXIT
# --- 0. preconditions ------------------------------------------------------
echo "==> checking preconditions"
psql_q "SELECT org_id FROM farms LIMIT 0" >/dev/null 2>&1 \
|| fail "no migrated database at DATABASE_URL - run scripts/dev-env.sh up && schema"
[ -x "$GO2RTC_BIN" ] || fail "no go2rtc at $GO2RTC_BIN (set GO2RTC_BIN)"
[ -f "$PROXY_DIR/printers.toml" ] || fail "no $PROXY_DIR/printers.toml"
grep -q "^id *= *\"$CAMERA_ID\"" "$PROXY_DIR/printers.toml" \
|| fail "printers.toml has no printer with id '$CAMERA_ID' (set CAMERA_ID)"
# A leftover go2rtc would answer for the one under test and make this pass or
# fail for reasons that have nothing to do with the code. Refuse to start.
if ps -eo comm | grep -qx go2rtc; then
fail "a go2rtc is already running - it would answer instead of the one under test: $(ps -eo pid,etime,cmd | awk '$3 ~ /go2rtc/' | head -2)"
fi
for p in 1984 8555; do
ss -ltnu 2>/dev/null | grep -q ":$p " && fail "port $p is already in use"
done
CAM_HOST="$(awk -v id="$CAMERA_ID" -F'"' '/^id *=/ {cur=$2} cur==id && /^(host|address|ip) *=/ {print $2; exit}' "$PROXY_DIR/printers.toml")"
if [ -n "$CAM_HOST" ]; then
timeout 4 bash -c "</dev/tcp/$CAM_HOST/322" 2>/dev/null \
|| fail "$CAMERA_ID ($CAM_HOST) refuses TCP 322: it has no RTSP camera, or is off"
ok "$CAMERA_ID accepts RTSP on 322"
else
ok "could not read $CAMERA_ID's address from printers.toml; skipping the 322 pre-check"
fi
# --- 1. seed ---------------------------------------------------------------
echo "==> seeding an organisation and a farm claiming '$CAMERA_ID'"
EDGE_PROXY_TOKEN="e2e-$(date +%s)"
ORG_ID="$(psql_q "INSERT INTO organisations (name) VALUES ('e2e camera org') RETURNING id")"
# `viewer`: the relay route asks for viewer, so seeding the weakest role that
# should work shows the camera is watchable without any authority to command.
psql_q "INSERT INTO org_members (org_id, user_id, role) VALUES ('$ORG_ID', '$USER_ID', 'viewer')" >/dev/null
FARM_ID="$(psql_q "INSERT INTO farms (name, org_id, edge_proxy_token)
VALUES ('e2e camera farm', '$ORG_ID', '$EDGE_PROXY_TOKEN') RETURNING id")"
psql_q "INSERT INTO printers (farm_id, name, vendor, gateway_printer_id)
VALUES ('$FARM_ID', 'camera under test', 'bambu', '$CAMERA_ID')" >/dev/null
CAM_UUID="$(psql_q "SELECT id FROM printers WHERE farm_id='$FARM_ID' AND gateway_printer_id='$CAMERA_ID'")"
ok "org $ORG_ID, farm $FARM_ID, viewer $USER_ID"
# --- 2. backend ------------------------------------------------------------
echo "==> starting the dev-auth issuer on :$AUTH_PORT and the backend on :$PORT"
bun scripts/dev-auth.ts --port "$AUTH_PORT" > "$WORK/auth.log" 2>&1 &
AUTH_PID=$!
AUTH_URL="http://localhost:$AUTH_PORT"
for _ in $(seq 1 40); do curl -sf "$AUTH_URL/auth/v1/.well-known/jwks.json" >/dev/null 2>&1 && break; sleep 0.25; done
curl -sf "$AUTH_URL/auth/v1/.well-known/jwks.json" >/dev/null 2>&1 || fail "dev-auth did not come up"
export SUPABASE_URL="$AUTH_URL"
USER_TOKEN="$(curl -sf "$AUTH_URL/token?sub=$USER_ID" | python3 -c 'import json,sys; print(json.load(sys.stdin)["access_token"])')"
bun run src/index.ts > "$WORK/backend.log" 2>&1 &
BACKEND_PID=$!
for _ in $(seq 1 60); do curl -sf "$BASE/health" >/dev/null 2>&1 && break; sleep 0.5; done
curl -sf "$BASE/health" >/dev/null || fail "backend did not become healthy"
TOKEN="$(curl -sf -X POST "$BASE/auth/edge-token" -H 'content-type: application/json' \
-d "{\"farmId\":\"$FARM_ID\",\"edgeProxyToken\":\"$EDGE_PROXY_TOKEN\"}" \
| python3 -c 'import json,sys; print(json.load(sys.stdin).get("token",""))')"
[ -n "$TOKEN" ] || fail "no edge token"
ok "backend healthy, edge token minted"
# --- 3. the real gateway, with the real go2rtc ------------------------------
echo "==> starting the real continuum-proxy, which spawns the real go2rtc"
( cd "$PROXY_DIR" && cargo build --quiet 2>&1 | tail -3 ) || fail "proxy did not build"
# STREAM_BASE_URL has to be set for /stream to answer 200; it is never
# dialled, because the media path here is the relay, not a gateway URL.
env -C "$PROXY_DIR" \
CONTINUUM_GATEWAY_ID=e2e-cam-gw \
CONTINUUM_FARM_ID="$FARM_ID" \
CONTINUUM_GATEWAY_TOKEN="$TOKEN" \
CONTINUUM_UPLINK_URL="ws://localhost:$PORT/ws/edge/v1" \
CONTINUUM_GO2RTC_BIN="$GO2RTC_BIN" \
CONTINUUM_GO2RTC_CONFIG="$WORK/go2rtc.yaml" \
CONTINUUM_STREAM_BASE_URL="http://e2e-gateway.invalid:1984" \
CONTINUUM_PRINTERS_PATH=printers.toml \
RUST_LOG=info \
"$PROXY_DIR/target/debug/continuum-proxy" > "$WORK/proxy.log" 2>&1 &
PROXY_PID=$!
echo "==> waiting for the camera to be advertised"
STREAM=""
for _ in $(seq 1 45); do
STREAM="$(curl -sf -H "authorization: Bearer $USER_TOKEN" "$BASE/printers/$CAM_UUID/stream" 2>/dev/null || true)"
[ -n "$STREAM" ] && break
sleep 1
done
[ -n "$STREAM" ] || fail "GET /printers/$CAM_UUID/stream never answered 200 - the gateway did not report a camera"
# Deliberately not echoed: with TURN configured the body carries a live
# username and credential, and printing it put them in a transcript once.
ok "the gateway advertises camera '$(printf '%s' "$STREAM" | python3 -c 'import json,sys; d=json.load(sys.stdin); print(d["cameraId"], "- iceServers:", len(d["iceServers"]))')'"
# The generated config holds the printer's access code and go2rtc's API
# password. $WORK is mode 0700, which hides a bad mode here, so the file's own
# mode is what is asserted - it was 0644 until this check caught it.
[ -f "$WORK/go2rtc.yaml" ] || fail "the proxy never wrote a go2rtc config"
mode="$(stat -c %a "$WORK/go2rtc.yaml")"
[ "$mode" = "600" ] || fail "go2rtc.yaml is mode $mode, but it holds an access code and must be 600"
ok "the generated go2rtc config is owner-only (mode $mode)"
# --- 4. the viewer ---------------------------------------------------------
echo "==> a real WebRTC peer plays the camera through the relay"
RESULTS="$(BASE="$BASE" PRINTER_UUID="$CAM_UUID" ACCESS_TOKEN="$USER_TOKEN" bun scripts/e2e-camera-peer.ts 2>&1 || true)"
SEEN=0; BAD=0
while IFS=$'\t' read -r verdict label; do
case "$verdict" in
ok) ok "$label"; SEEN=$((SEEN + 1)) ;;
FAILED) echo " FAILED: $label" >&2; SEEN=$((SEEN + 1)); BAD=$((BAD + 1)) ;;
esac
done < <(printf '%s\n' "$RESULTS" | grep -E "^(ok|FAILED)$(printf '\t')")
# Fewer results than the peer's five checks means it died before finishing,
# and a silent peer must not read as a pass.
if [ "$SEEN" -lt 5 ] && [ "$BAD" -eq 0 ]; then
BAD=1; echo " FAILED: the peer produced only $SEEN of 5 results" >&2
fi
if [ "$BAD" -ne 0 ]; then
printf '%s\n' "$RESULTS" | sed 's/^/ | /' >&2
fail "the camera did not deliver video"
fi
echo
echo "PASS - a real WebRTC peer received video from the printer's camera"
echo " through the backend relay, the uplink, the proxy and go2rtc."
+5 -1
View File
@@ -47,6 +47,10 @@ R2_SECRET_ACCESS_KEY="${R2_SECRET_ACCESS_KEY:-e2e}"
R2_ENDPOINT="${R2_ENDPOINT:-http://localhost:9000}" R2_ENDPOINT="${R2_ENDPOINT:-http://localhost:9000}"
export DATABASE_URL VALKEY_URL EDGE_JWT_SECRET PORT export DATABASE_URL VALKEY_URL EDGE_JWT_SECRET PORT
export R2_ACCESS_KEY_ID R2_SECRET_ACCESS_KEY R2_ENDPOINT export R2_ACCESS_KEY_ID R2_SECRET_ACCESS_KEY R2_ENDPOINT
# This script asserts TURN is unconfigured (below). Sourcing dev-secrets.sh
# turns it on, which used to fail the assertion - and print the live
# credentials in the failure message. Make the assumption true.
unset CLOUDFLARE_TURN_KEY_ID CLOUDFLARE_TURN_API_TOKEN
WORK="$(mktemp -d)" WORK="$(mktemp -d)"
BACKEND_PID="" BACKEND_PID=""
@@ -255,7 +259,7 @@ fi
ok "every printer id this gateway reported mapped to a row" ok "every printer id this gateway reported mapped to a row"
echo "==> what landed in the database:" echo "==> what landed in the database:"
PGPASSWORD=continuum psql -h localhost -U continuum -d continuum -c \ psql "$DATABASE_URL" -c \
"SELECT gateway_printer_id AS gw, status, vendor_state, progress_percent AS pct, "SELECT gateway_printer_id AS gw, status, vendor_state, progress_percent AS pct,
current_layer || '/' || total_layers AS layers, remaining_minutes AS mins, current_layer || '/' || total_layers AS layers, remaining_minutes AS mins,
nozzle_temp_c AS nozzle, bed_temp_c AS bed, active_alerts AS alerts, nozzle_temp_c AS nozzle, bed_temp_c AS bed, active_alerts AS alerts,