deploying-hy2-nodes

TunPilot Node Deployment (Production-Optimized)

Deploy a production-grade Hysteria2 proxy node with automatic performance tuning, security hardening, and censorship resistance. Follow each phase in order.

Prerequisite: TunPilot server must be running and CLI must be configured (use getting-started skill if not).

---

Phase 1: Gather Information & Probe Server

1.1 Ask the User

Collect the following from the user:

• SSH destination: e.g. [email protected]
• Domain name (optional): A domain pointing to this server's IP. If none, self-signed certs will be used.
• Node name: A human-readable label (e.g. tokyo-01, bwg-us)

1.2 Test SSH Connectivity

ssh <server> "echo ok"

1.3 Probe Server Capabilities

Run ALL probes in a single SSH session to minimize round trips:

ssh <server> bash <<'PROBE'
echo "=== OS/ARCH ==="
uname -s -m
cat /etc/os-release 2>/dev/null | grep -E '^(ID|VERSION_ID)='

echo "=== CPU ==="
nproc

echo "=== MEMORY ==="
free -b | awk '/Mem/{print $2}'

echo "=== PORT CONFLICTS ==="
ss -tulnp | grep -E ':443|:80' || echo "no conflicts"

echo "=== FIREWALL ==="
if command -v ufw &>/dev/null; then echo "ufw"; ufw status 2>/dev/null
elif command -v firewall-cmd &>/dev/null; then echo "firewalld"; firewall-cmd --state 2>/dev/null
elif command -v nft &>/dev/null; then echo "nftables"
else echo "none"
fi

echo "=== EXISTING HY2 ==="
hysteria version 2>/dev/null || echo "not installed"

echo "=== NETWORK ==="
ip -4 addr show scope global 2>/dev/null
ip -6 addr show scope global 2>/dev/null

echo "=== SYSCTL ==="
sysctl -n net.core.rmem_max 2>/dev/null
sysctl -n net.core.wmem_max 2>/dev/null
sysctl -n net.ipv4.tcp_congestion_control 2>/dev/null
sysctl -n net.core.default_qdisc 2>/dev/null
PROBE

1.4 Build Server Profile

Using the probe results, build a server profile table:

Parameter	Source	Derived Setting
Memory	free -b	QUIC receive/send window size (Memory < 4 GB: 8 MB windows, Memory >= 4 GB: 16 MB windows)
CPU cores	nproc	maxStreams (cores x 256, cap at 1024)
Port conflicts	ss -tulnp	Whether to use alternative ports
Firewall type	probe	Which firewall commands to use (ufw/firewall-cmd/iptables/none)
Kernel tuning	sysctl values	Whether sysctl tuning is needed
IPv6	ip -6 addr	Whether to enable dual-stack

1.5 Confirm Choices with User

Present the server profile and confirm:

• Congestion control: Brutal (recommended for dedicated bandwidth) vs BBR (for shared/variable bandwidth)
• Bandwidth limits: Up/down values based on server specs
• Masquerade site: Default https://www.bing.com/ or custom

---

Phase 2: Deploy

2.1 Kernel Tuning

Apply QUIC-optimized sysctl settings. Skip if the probe shows values are already tuned.

ssh <server> bash <<'SYSCTL'
cat > /etc/sysctl.d/99-hysteria.conf << 'EOF'
# QUIC/UDP buffer sizes (official Hysteria2 recommendation)
net.core.rmem_max = 16777216
net.core.wmem_max = 16777216

# Queueing discipline (supports pacing needed by BBR)
net.core.default_qdisc = fq
net.ipv4.tcp_congestion_control = bbr
EOF

sysctl -p /etc/sysctl.d/99-hysteria.conf
SYSCTL

2.2 Install / Upgrade Hysteria2

ssh <server> "bash <(curl -fsSL https://get.hy2.sh/)"

Verify installation:

ssh <server> "hysteria version"

2.3 Install Diagnostic Dependencies

Install tools required by the diagnostic scripts (IPQuality + NetQuality):

ssh <server> "apt-get update -qq && apt-get install -y -qq jq curl bc netcat-openbsd dnsutils iproute2 iperf3 mtr"

These are needed for the testing-nodes skill diagnostics (IPQuality + NetQuality) to work without prompting for interactive installation. NetQuality's remaining dependencies (speedtest, nexttrace) are auto-installed by the script's -y flag on first run.

2.3.1 Install Diagnostics Wrapper

Deploy the tunpilot-diag script for clean JSON diagnostics output:

ssh <server> bash <<'DIAG_INSTALL'
curl -fsSL https://raw.githubusercontent.com/Buywatermelon/tunpilot/main/scripts/tunpilot-diag.sh \
  -o /usr/local/bin/tunpilot-diag
chmod +x /usr/local/bin/tunpilot-diag
tunpilot-diag --version
DIAG_INSTALL

2.4 TLS Certificate

Option A — With domain (ACME handled by Hysteria2 config):

ACME is configured directly in the Hysteria2 config.yaml (see Phase 2.6). There is no separate hysteria cert command needed. Just ensure port 80 is open for the HTTP-01 challenge:

ssh <server> bash <<'ACME_PREP'
# Ensure port 80 is not occupied by another service
ss -tlnp | grep ':80 ' && echo "WARNING: port 80 in use — ACME may fail" || echo "port 80 available"
mkdir -p /etc/hysteria
ACME_PREP

Option B — Without domain (self-signed EC P-256):

ssh <server> bash <<'SELFSIGN'
mkdir -p /etc/hysteria
openssl req -x509 -newkey ec \
  -pkeyopt ec_paramgen_curve:prime256v1 \
  -keyout /etc/hysteria/key.pem \
  -out /etc/hysteria/cert.pem \
  -days 3650 -nodes \
  -subj '/CN=bing.com'
SELFSIGN

2.5 Register Node in TunPilot

Use the CLI to register the node:

tunpilot node add \
  --name=<node-name> \
  --host=<server-ip-or-domain> \
  --port=443 \
  --protocol=hysteria2 \
  --stats_port=9999 \
  --stats_secret=<random-string> \
  --sni=<domain> \
  --insecure=1

Required flags: --name, --host, --port, --protocol

Optional flags: --stats_port, --stats_secret, --sni, --cert_path, --ssh_user, --ssh_port, --insecure (1 for self-signed, 0 for ACME)

Save the returned node ID and auth_callback_url from the JSON output.

2.6 Write Production Config

Read the config template from hysteria2-template.md in this skill directory. Choose the appropriate config variant:

• Config A (ACME / with domain): Uses acme block for automatic certificate management
• Config B (Self-signed / no domain): Uses cert and key paths directly

Fill all placeholders using values from the server profile built in Phase 1.4. Write the config:

ssh <server> "cat > /etc/hysteria/config.yaml << 'CONF'
<filled config from template>
CONF"

Adjust bandwidth based on the server's actual network capacity and the user's confirmed choices from Phase 1.5.

2.7 Systemd Hardening

Create a systemd drop-in to harden the Hysteria2 service:

ssh <server> bash <<'SYSTEMD'
mkdir -p /etc/systemd/system/hysteria-server.service.d

cat > /etc/systemd/system/hysteria-server.service.d/hardening.conf << 'EOF'
[Service]
LimitNOFILE=65536
NoNewPrivileges=true
AmbientCapabilities=CAP_NET_ADMIN CAP_NET_BIND_SERVICE CAP_NET_RAW
CapabilityBoundingSet=CAP_NET_ADMIN CAP_NET_BIND_SERVICE CAP_NET_RAW
ProtectSystem=strict
ProtectHome=true
ReadWritePaths=/etc/hysteria
EOF

systemctl daemon-reload
SYSTEMD

2.8 Firewall

Open required ports using the firewall type detected in Phase 1.3:

ssh <server> bash <<'FIREWALL'
# Detect and apply firewall rules
if command -v ufw &>/dev/null; then
  ufw allow 443/udp
  ufw allow 443/tcp
  ufw allow 80/tcp
  ufw reload
elif command -v firewall-cmd &>/dev/null; then
  firewall-cmd --permanent --add-port=443/udp
  firewall-cmd --permanent --add-port=443/tcp
  firewall-cmd --permanent --add-port=80/tcp
  firewall-cmd --reload
else
  echo "No firewall manager detected — ensure UDP/443, TCP/443, TCP/80 are open at the provider level"
fi
FIREWALL

2.9 Start Service

ssh <server> "systemctl enable --now hysteria-server && sleep 2 && systemctl is-active hysteria-server"

If the service fails to start, check logs immediately:

ssh <server> "journalctl -u hysteria-server --no-pager -n 50"

---

Phase 3: Verify

3.1 Health Check

Use the CLI to check node health:

tunpilot health

3.2 Masquerade Test

Only if a domain was configured — verify the masquerade proxy is working:

curl -I https://<domain>

The response should show headers from the masquerade target (e.g. Bing).

3.3 Stats API Test

Test the traffic stats API from the node itself via SSH:

ssh <server> "curl -s -H 'Authorization: <stats_secret>' http://127.0.0.1:9999/online"

This should return a JSON response with online user count.

3.4 Log Check

Review recent logs for any errors or warnings:

ssh <server> "journalctl -u hysteria-server --no-pager -n 30 --since '5 minutes ago'"

3.5 Deployment Summary

Present a final report to the user:

• Node name and ID
• Server IP and domain (if any)
• Protocol and port
• TLS type (ACME or self-signed)
• Congestion control and bandwidth limits
• Kernel tuning applied
• Health check result
• Subscription instructions (use tunpilot user update <id> --nodes=<node-id> to grant users access)

---

Troubleshooting

Symptom	Diagnosis	Fix
tunpilot health unreachable	Stats API not accessible	Verify stats_port and stats_secret match between TunPilot and the node config
Service won't start	Config syntax error	Run journalctl -u hysteria-server --no-pager -n 50 and validate YAML syntax
ACME cert fails	DNS not pointing to server	Check dig <domain>, ensure port 80 is open and not occupied
Clients can't connect	Firewall blocking UDP/443	Check `ss -ulnp
Slow speeds	Wrong congestion control	Check Brutal bandwidth setting matches actual capacity, try switching to BBR
Auth failures	Callback URL unreachable	Run curl <auth_callback_url> from the node to verify TunPilot is reachable

---

CLI Commands Reference

Command	Use When
tunpilot node list	See all registered nodes
tunpilot node add --name=... --host=... --port=... --protocol=...	Register a new node (Phase 2.5)
tunpilot node update <id> --name=...	Change node config (port, SNI, enable/disable)
tunpilot node remove <id>	Delete a node (cascades user assignments)
tunpilot node sync	Sync all node configurations
tunpilot health	Verify all nodes are reachable
tunpilot traffic --node=<id>	Query traffic usage by node
tunpilot user update <id> --nodes=<id1>,<id2>	Grant a user access to specific nodes
tunpilot sub create --user=<id> --format=surge	Generate client subscription link for a user