talk: improve startup ordering, NATS tuning, signaling timeouts, and healthcheck depth

Agent-Logs-Url: https://github.com/nextcloud/all-in-one/sessions/7fd39619-0f63-494c-902c-746f00d17d02
Co-Authored-By: szaimen <42591237+szaimen@users.noreply.github.com>
Signed-off-by: Simon L. <szaimen@e.mail.de>

Containers/talk/Dockerfile

@@ -82,7 +82,9 @@ RUN set -ex; \
     touch \
         /etc/nats.conf \
         /etc/eturnal.yml; \
-    echo "listen: 127.0.0.1:4222" | tee /etc/nats.conf; \
+# write_deadline: "10s" — without a write deadline, a lagging subscriber can stall the broker indefinitely, blocking all other signaling messages.
+# max_payload: 8MB — the default is 1 MB; signaling payloads in large meetings (many participants, ICE candidates) can exceed this, causing dropped messages.
+    printf 'listen: 127.0.0.1:4222\nwrite_deadline: "10s"\nmax_payload: 8MB\n' | tee /etc/nats.conf; \
     mkdir -p \
         /var/tmp \
         /conf \

Containers/talk/healthcheck.sh

@@ -5,3 +5,6 @@ nc -z 127.0.0.1 8188 || exit 1
 nc -z 127.0.0.1 4222 || exit 1
 nc -z 127.0.0.1 "$TALK_PORT" || exit 1
 eturnalctl status || exit 1
+# Verify that the signaling server is actually serving requests, not just
+# listening on the TCP port (which nc -z above only tests for open port).
+wget -q -O /dev/null http://127.0.0.1:8081/api/v1/stats || exit 1

Containers/talk/start.sh

@@ -91,10 +91,12 @@ if [ -z "$TALK_MAX_SCREEN_BITRATE" ]; then
     TALK_MAX_SCREEN_BITRATE=2097152
 fi
 
-# Signling
+# Signaling
 cat << SIGNALING_CONF > "/conf/signaling.conf"
 [http]
 listen = 0.0.0.0:8081
+readtimeout = 15
+writetimeout = 30
 
 [app]
 debug = false
@@ -110,7 +112,9 @@ internalsecret = ${INTERNAL_SECRET}
 backends = backend-1
 allowall = false
 timeout = 10
-connectionsperhost = 8
+# connectionsperhost: This is the HTTP keep-alive connection pool size from the signaling server to the Nextcloud backend. 
+# Under load (many concurrent calls joining/leaving simultaneously) a pool of 8 creates a queue bottleneck for backend authentication and session lookups, thus increasing to 32.
+connectionsperhost = 32
 skipverify = ${SKIP_CERT_VERIFY}
 
 [backend-1]

Containers/talk/supervisord.conf

@@ -7,19 +7,23 @@ logfile_maxbytes=50MB
 logfile_backups=10                              
 loglevel=error
 
-[program:eturnal]
-stdout_logfile=/dev/stdout
-stdout_logfile_maxbytes=0
-stderr_logfile=/dev/stderr
-stderr_logfile_maxbytes=0
-command=eturnalctl foreground
-
 [program:nats-server]
 stdout_logfile=/dev/stdout
 stdout_logfile_maxbytes=0
 stderr_logfile=/dev/stderr
 stderr_logfile_maxbytes=0
 command=nats-server -c /etc/nats.conf
+# Start first: signaling depends on NATS being available
+priority=10
+
+[program:eturnal]
+stdout_logfile=/dev/stdout
+stdout_logfile_maxbytes=0
+stderr_logfile=/dev/stderr
+stderr_logfile_maxbytes=0
+command=eturnalctl foreground
+# Start alongside Janus; independent of signaling
+priority=20
 
 [program:janus]
 stdout_logfile=/dev/stdout
@@ -28,6 +32,8 @@ stderr_logfile=/dev/stderr
 stderr_logfile_maxbytes=0
 # debug-level 3 means warning
 command=janus --config=/conf/janus.jcfg --disable-colors --log-stdout --full-trickle --debug-level 3
+# Start alongside eturnal; signaling connects to Janus via WebSocket
+priority=20
 
 [program:signaling]
 stdout_logfile=/dev/stdout
@@ -35,3 +41,5 @@ stdout_logfile_maxbytes=0
 stderr_logfile=/dev/stderr
 stderr_logfile_maxbytes=0
 command=nextcloud-spreed-signaling -config /conf/signaling.conf
+# Start last: depends on NATS (priority=10) and Janus (priority=20) being up
+priority=30