feat(cluster0): add Janus SFU deploy script and graph gzip smoke test

deploy-janus-sfu.sh installs Janus on cast02, writes signaling config, and templates CAST02_IP into nginx; populate gains --phase sfu.

Co-authored-by: Cursor <cursoragent@cursor.com>

sim/cluster0/lab-seeds/backend/scripts/test_graphs_api.sh

@@ -76,7 +76,24 @@ echo
 echo "…"
 
 if [[ "${upload_code}" != "200" ]]; then
-  echo "FAIL: upload expected 200" >&2
+  echo "FAIL: plain upload expected 200" >&2
+  exit 1
+fi
+
+sid_gzip="smoke-graph-gzip-$(date +%s)"
+payload_gzip="${payload/${sid}/${sid_gzip}}"
+echo "POST ${UPLOAD_URL} (gzip, device-style)"
+gzip_upload_code="$(curl -sS -o /tmp/graph_upload_gzip_resp.json -w '%{http_code}' \
+  -H 'Content-Type: application/json' \
+  -H 'Content-Encoding: gzip' \
+  --data-binary @<(printf '%s' "${payload_gzip}" | gzip -c) \
+  "${UPLOAD_URL}")"
+echo "gzip upload HTTP ${gzip_upload_code}"
+cat /tmp/graph_upload_gzip_resp.json
+echo
+
+if [[ "${gzip_upload_code}" != "200" ]]; then
+  echo "FAIL: gzip upload expected 200" >&2
   exit 1
 fi
 if [[ "${graphs_code}" != "200" ]]; then

sim/cluster0/nginx/apps-port80.conf

@@ -290,7 +290,7 @@ server {
 
 	# WebSocket → Janus on cast02
 	location /app/androidcast_project/sfu/ws {
-		proxy_pass         http://10.7.16.237:8188/;
+		proxy_pass         http://@CAST02_IP@:8188/;
 		proxy_http_version 1.1;
 		proxy_set_header   Upgrade $http_upgrade;
 		proxy_set_header   Connection $connection_upgrade;

sim/cluster0/populate_lab_setup.sh

@@ -3,7 +3,7 @@
 # Usage:
 #   sudo sh populate_lab_setup.sh              # full setup on this node
 #   sudo sh populate_lab_setup.sh --coordinator # from jump host: all nodes in order
-#   sudo sh populate_lab_setup.sh --phase baseline|db|app|all
+#   sudo sh populate_lab_setup.sh --phase baseline|db|app|sfu|all
 set -eu
 
 ROOT="$(cd "$(dirname "$0")" && pwd)"
@@ -20,7 +20,7 @@ while [ $# -gt 0 ]; do
     --coordinator) COORDINATOR=1; shift ;;
     --phase) PHASE="${2:-all}"; shift 2 ;;
     --help|-h)
-      echo "Usage: sudo sh populate_lab_setup.sh [--coordinator] [--phase baseline|db|app|all]"
+      echo "Usage: sudo sh populate_lab_setup.sh [--coordinator] [--phase baseline|db|app|sfu|all]"
       exit 0
       ;;
     *) die "unknown arg: $1" ;;
@@ -61,6 +61,18 @@ run_phase_db() {
   die "unknown host for db phase: $(host_short)"
 }
 
+run_phase_sfu() {
+  JOURNAL_PHASE=sfu
+  journal_start phase_sfu "$(host_short)"
+  log "phase sfu on $(host_short)"
+  if is_replica_node; then
+    wait_for_primary_db
+  fi
+  run_script "$ROOT/scripts/deploy-ac-workspace.sh"
+  run_script "$ROOT/scripts/deploy-janus-sfu.sh"
+  journal_ok phase_sfu
+}
+
 run_phase_app() {
   JOURNAL_PHASE=app
   journal_start phase_app "$(host_short)"
@@ -109,6 +121,7 @@ run_local() {
     baseline) run_phase_baseline ;;
     db) run_phase_baseline; run_phase_db ;;
     app) run_phase_app ;;
+    sfu) run_phase_sfu ;;
     all)
       run_phase_baseline || exit 1
       run_phase_db || exit 1
@@ -142,6 +155,12 @@ run_coordinator() {
   done
   wait
 
+  ssh_node "$CAST02_HOST" "sudo sh ${SHARED_MOUNT}/cluster/populate_lab_setup.sh --phase sfu"
+  for H in $CAST01_HOST $CAST03_HOST; do
+    ssh_node "$H" "sudo sh ${SHARED_MOUNT}/cluster/populate_lab_setup.sh --phase sfu" &
+  done
+  wait
+
   sh "$ROOT/verify_lab_setup.sh" --cluster
   journal_ok populate_coordinator
   log "populate_coordinator_ok ${CLUSTER_NAME}"

sim/cluster0/scripts/configure-nginx.sh

@@ -17,7 +17,11 @@ if [ -f "$GIT_NGINX" ] && [ "$(realpath "$GIT_NGINX" 2>/dev/null)" != "$(realpat
   fi
 fi
 [ -s "$CONF_DST" ] || die "missing or empty nginx template at $CONF_DST"
-install -D -m 644 "$CONF_DST" /etc/nginx/http.d/androidcast.conf
+load_cluster_env 2>/dev/null || true
+CAST02_IP="${CAST02_IP:-10.7.16.237}"
+sed "s/@CAST02_IP@/${CAST02_IP}/g" "$CONF_DST" > /tmp/androidcast.conf.$$
+install -D -m 644 /tmp/androidcast.conf.$$ /etc/nginx/http.d/androidcast.conf
+rm -f /tmp/androidcast.conf.$$
 rm -f /etc/nginx/http.d/default.conf /etc/nginx/http.d/cluster.conf 2>/dev/null || true
 nginx -t
 rc-service nginx reload

sim/cluster0/scripts/deploy-janus-sfu.sh

@@ -0,0 +1,83 @@
+#!/bin/sh
+# Janus SFU on cast02 + ac-ms-sfu-signaling config on all app nodes.
+set -eu
+ROOT="$(cd "$(dirname "$0")/.." && pwd)"
+# shellcheck source=/dev/null
+. "$ROOT/scripts/lib/common.sh"
+load_cluster_env
+
+WS="${AC_WORKSPACE_ROOT:-/var/www/ac/workspace}"
+AC_REPO="${AC_MONOREPO_ROOT:-/mnt/repos/ac}"
+SFU_WS="${WS}/ac-ms-sfu-signaling"
+if [ ! -d "$SFU_WS" ] && [ -d "${AC_REPO}/ac-ms-sfu-signaling" ]; then
+  mkdir -p "$(dirname "$SFU_WS")"
+  rsync -a --delete "${AC_REPO}/ac-ms-sfu-signaling/" "$SFU_WS/" 2>/dev/null \
+    || cp -a "${AC_REPO}/ac-ms-sfu-signaling/." "$SFU_WS/"
+fi
+[ -d "$SFU_WS" ] || die "missing ac-ms-sfu-signaling (run deploy-ac-workspace.sh first)"
+
+PUBLIC_ORIGIN="${LAB_PUBLIC_ORIGIN:-https://apps.f0xx.org}"
+JANUS_HTTP="http://${CAST02_IP}:8088/janus"
+JANUS_WS="ws://${CAST02_IP}:8188"
+
+install_janus_on_cast02() {
+  [ "$(host_short)" = "$CAST02_HOST" ] || return 0
+  log "cast02: install/configure Janus SFU"
+  if ! command -v janus >/dev/null 2>&1; then
+    apk add --no-cache janus-gateway janus-gateway-openrc
+  fi
+  for _pair in \
+    "janus.jcfg janus.jcfg.sample" \
+    "janus.transport.http.jcfg janus.transport.http.jcfg.sample" \
+    "janus.transport.websockets.jcfg janus.transport.websockets.jcfg.sample" \
+    "janus.plugin.videoroom.jcfg janus.plugin.videoroom.jcfg.sample"
+  do
+    _dst="${_pair%% *}"
+    _src="${_pair#* }"
+    [ -f "/etc/janus/${_dst}" ] || cp "/etc/janus/${_src}" "/etc/janus/${_dst}"
+  done
+  rc-update add janus-gateway default 2>/dev/null || true
+  rc-service janus-gateway restart 2>/dev/null || rc-service janus-gateway start 2>/dev/null || true
+  if curl -sf "http://127.0.0.1:8088/janus/info" >/dev/null 2>&1; then
+    log "cast02: Janus HTTP API OK"
+  else
+    log "WARN cast02: Janus health check failed (may still be starting)"
+  fi
+}
+
+write_sfu_config() {
+  DB_PASS="$(read_cred mariadb_app password 2>/dev/null || echo 'CHANGE_ME')"
+  mkdir -p "${SFU_WS}/config"
+  cat > "${SFU_WS}/config/config.php" <<PHPEOF
+<?php
+return [
+    'janus_url'     => '${JANUS_HTTP}',
+    'janus_ws_url'  => '${JANUS_WS}',
+    'signaling_url' => '${PUBLIC_ORIGIN}/app/androidcast_project/sfu/ws',
+    'db_dsn'        => 'mysql:host=${PRIMARY_DB_HOST};dbname=androidcast_crashes;charset=utf8mb4',
+    'db_user'       => 'androidcast',
+    'db_pass'       => '${DB_PASS}',
+];
+PHPEOF
+  chown -R nginx:nginx "$SFU_WS" 2>/dev/null || true
+  log "sfu signaling config → ${SFU_WS}/config/config.php"
+}
+
+apply_sfu_schema() {
+  is_primary_node || return 0
+  log "primary: apply SFU schema migrations"
+  run_script "$ROOT/scripts/load-schemas.sh" || log "WARN load-schemas returned non-zero (may already be applied)"
+}
+
+install_janus_on_cast02
+write_sfu_config
+apply_sfu_schema
+run_script "$ROOT/scripts/configure-nginx.sh"
+rc-service php-fpm83 restart 2>/dev/null || true
+
+if [ "$(host_short)" = "$CAST02_HOST" ]; then
+  curl -sf "http://127.0.0.1:8088/janus/info" | head -c 120 || true
+  echo
+fi
+
+log "deploy-janus-sfu_ok $(host_short)"