Vertical scaling (same server)
Add more workers
Each worker handles one call at a time. To increase capacity, add more worker instances:
- Create new systemd instances and add sockets to nginx:
# Enable workers 17-20
for i in $(seq 17 20); do
systemctl enable "cxbcore@$i"
systemctl start "cxbcore@$i"
done
- Add new sockets to nginx upstream:
upstream cxbcore_workers {
least_conn;
server unix:/tmp/cxbcore_1.sock max_conns=1 fail_timeout=10;
...
server unix:/tmp/cxbcore_16.sock max_conns=1 fail_timeout=10;
# New workers
server unix:/tmp/cxbcore_17.sock max_conns=1 fail_timeout=10;
server unix:/tmp/cxbcore_18.sock max_conns=1 fail_timeout=10;
server unix:/tmp/cxbcore_19.sock max_conns=1 fail_timeout=10;
server unix:/tmp/cxbcore_20.sock max_conns=1 fail_timeout=10;
}
- Reload nginx and update
.env:
nginx -t && systemctl reload nginx
# Update CXBCORE_WORKERS in .env to match (used by health endpoint)
Memory budget: each idle worker uses ~260 MB. On an 8 GB server, 16 workers is the safe maximum (4 GB baseline + headroom for active calls). Add more RAM before adding more workers.
Horizontal scaling (multiple servers)
Add a new fleet server
Adding a fleet touches two independent paths — inbound and outbound are wired separately. Do both, or the new fleet only serves half of your traffic.
Deploy CXB Core
Set up CXB Core on the new server with the same code, .env, and systemd template. Start 16 workers. Confirm https://<new-host>/health/fleet responds before continuing.
Configure fleet nginx
Set up nginx with the same Unix socket upstream pattern (max_conns=1 per socket), the 429-retry blocks for /attach, /livekit/dialout, /livekit/widget, and SSL.
(1) Inbound — register with the WSS ingress
If inbound WebSocket calls (telephony dialler, Exotel) are fronted by the HAProxy ingress, run add-fleet.sh <new-host> on the ingress host. It clones the last server line, validates with haproxy -c, reloads zero-downtime, and rolls back automatically if the new backend does not come UP. Do not hand-edit HAProxy server lines. See WSS ingress. (2) Outbound — register in CXB API
Add the new server URL to CXB API system settings (fleet list in the CXB Console Settings page) so outbound calls are distributed to it. The fleet picker health-checks the list on every dialout and picks it up immediately — no restart, no code change.
The HAProxy WSS ingress is a single point of failure until a second ingress + managed LB/VRRP is added. Plan the second ingress before crossing ~3 fleets. The ingress is WSS-only, so the CXB API fleet registration (path 2) is what makes a new fleet usable for outbound calls regardless of the ingress.
Zero-downtime restarts
Restart workers one at a time. While one restarts, the other 15 continue serving:
for i in $(seq 1 16); do
systemctl restart "cxbcore@$i"
sleep 5
done
