lightscale/lab/test-nat-churn.nix
centra 559d6d2c6e
Some checks are pending
linux-lab / core-fast (push) Waiting to run
linux-lab / core-nat (push) Waiting to run
linux-lab / core-netem (push) Waiting to run
linux-lab / extended-auth-url (push) Waiting to run
linux-lab / extended-nat-churn (push) Waiting to run
linux-lab / extended-relay-switch (push) Waiting to run
linux-lab / extended-soak (push) Waiting to run
linux-lab / extended-standalone (push) Waiting to run
Switch submodule remotes to GitHub mirrors
2026-02-24 02:41:01 +09:00

215 lines
8.1 KiB
Nix

{ pkgs, serverPkg, clientPkg }:
{
name = "lightscale-lab-nat-churn";
nodes = {
node1 = { ... }: {
networking.hostName = "node1";
networking.usePredictableInterfaceNames = false;
virtualisation.vlans = [ 1 ];
networking.interfaces.eth1.useDHCP = false;
networking.interfaces.eth1.ipv4.addresses = [
{ address = "10.0.0.1"; prefixLength = 24; }
];
networking.firewall.enable = false;
boot.kernelModules = [ "wireguard" ];
environment.systemPackages = [
serverPkg
clientPkg
pkgs.wireguard-tools
pkgs.iproute2
pkgs.iputils
pkgs.netcat-openbsd
pkgs.curl
];
};
natgw = { ... }: {
networking.hostName = "natgw";
networking.usePredictableInterfaceNames = false;
virtualisation.vlans = [ 1 2 ];
networking.interfaces.eth1.useDHCP = false;
networking.interfaces.eth1.ipv4.addresses = [
{ address = "10.0.0.2"; prefixLength = 24; }
];
networking.interfaces.eth2.useDHCP = false;
networking.interfaces.eth2.ipv4.addresses = [
{ address = "192.168.60.1"; prefixLength = 24; }
];
networking.firewall.enable = false;
boot.kernel.sysctl."net.ipv4.ip_forward" = 1;
environment.systemPackages = [
pkgs.iproute2
pkgs.iputils
pkgs.iptables
pkgs.conntrack-tools
];
};
node3 = { ... }: {
networking.hostName = "node3";
networking.usePredictableInterfaceNames = false;
virtualisation.vlans = [ 2 ];
networking.interfaces.eth1.useDHCP = false;
networking.interfaces.eth1.ipv4.addresses = [
{ address = "192.168.60.10"; prefixLength = 24; }
];
networking.defaultGateway = {
address = "192.168.60.1";
interface = "eth1";
};
networking.firewall.enable = false;
boot.kernelModules = [ "wireguard" ];
environment.systemPackages = [
clientPkg
pkgs.wireguard-tools
pkgs.iproute2
pkgs.iputils
pkgs.curl
];
};
};
testScript = ''
start_all()
node1.wait_for_unit("multi-user.target")
natgw.wait_for_unit("multi-user.target")
node3.wait_for_unit("multi-user.target")
node1.wait_until_succeeds("ip -4 addr show dev eth1 | grep -q '10.0.0.1/24'")
natgw.wait_until_succeeds("ip -4 addr show dev eth1 | grep -q '10.0.0.2/24'")
natgw.wait_until_succeeds("ip -4 addr show dev eth2 | grep -q '192.168.60.1/24'")
node3.wait_until_succeeds("ip -4 addr show dev eth1 | grep -q '192.168.60.10/24'")
natgw.succeed("iptables -P FORWARD ACCEPT")
natgw.succeed("iptables -F")
natgw.succeed("iptables -t nat -F")
def set_nat_port(port):
natgw.succeed("iptables -t nat -F")
natgw.succeed(
f"iptables -t nat -A PREROUTING -i eth1 -p udp --dport {port} "
"-j DNAT --to-destination 192.168.60.10:51820"
)
natgw.succeed(
f"iptables -t nat -A POSTROUTING -o eth1 -p udp -s 192.168.60.10 "
f"--sport 51820 -j SNAT --to-source 10.0.0.2:{port}"
)
natgw.succeed("iptables -t nat -A POSTROUTING -o eth1 -j MASQUERADE")
natgw.execute("conntrack -F || true")
set_nat_port(40000)
# Very short UDP conntrack timeout to emulate consumer NAT churn.
natgw.succeed("sysctl -w net.netfilter.nf_conntrack_udp_timeout=5")
natgw.succeed("sysctl -w net.netfilter.nf_conntrack_udp_timeout_stream=5")
node1.succeed("touch /tmp/lightscale-server.log")
node1.execute("sh -c 'tail -n +1 -f /tmp/lightscale-server.log >/dev/console 2>&1 &'")
node1.succeed(
"systemd-run --no-block --unit=lightscale-server --service-type=simple "
"--property=Restart=on-failure --property=RestartSec=1 "
"--property=TimeoutStartSec=30 "
"--property=StandardOutput=append:/tmp/lightscale-server.log "
"--property=StandardError=append:/tmp/lightscale-server.log "
"--setenv=RUST_LOG=info --setenv=LIGHTSCALE_ADMIN_TOKEN=test-admin -- "
"lightscale-server --listen 10.0.0.1:8080 --state /tmp/lightscale-state.json"
)
node1.wait_for_unit("lightscale-server.service")
node1.wait_for_open_port(8080, addr="10.0.0.1", timeout=120)
import json
net = json.loads(node1.succeed(
"curl -sSf -X POST http://10.0.0.1:8080/v1/networks "
"-H 'authorization: Bearer test-admin' "
"-H 'content-type: application/json' "
"-d '{\"name\":\"nat-churn\",\"bootstrap_token_ttl_seconds\":600,"
"\"bootstrap_token_uses\":10,\"bootstrap_token_tags\":[\"nat-churn\"]}'"
))
token = net["bootstrap_token"]["token"]
def enroll(node, name, endpoints):
node.succeed(
"lightscale-client --profile natchurn --config /tmp/ls-config.json "
"init http://10.0.0.1:8080"
)
node.succeed(
f"lightscale-client --profile natchurn --config /tmp/ls-config.json "
f"--state-dir /tmp/ls-state register --node-name {name} -- {token}"
)
cmd = (
"lightscale-client --profile natchurn --config /tmp/ls-config.json "
"--state-dir /tmp/ls-state heartbeat"
)
for endpoint in endpoints:
cmd += f" --endpoint {endpoint}"
node.succeed(cmd)
enroll(node1, "node1", ["10.0.0.1:51820"])
enroll(node3, "node3", ["10.0.0.2:40000", "10.0.0.2:41000", "10.0.0.2:42000", "10.0.0.2:43000"])
def start_agent(node, endpoints):
node.succeed("touch /tmp/lightscale-agent.log")
cmd = (
"lightscale-client --profile natchurn --config /tmp/ls-config.json "
"--state-dir /tmp/ls-state agent --listen-port 51820 "
"--heartbeat-interval 5 --longpoll-timeout 5 "
"--endpoint-stale-after 5 --endpoint-max-rotations 2"
)
for endpoint in endpoints:
cmd += f" --endpoint {endpoint}"
node.succeed(
"systemd-run --no-block --unit=lightscale-agent --service-type=simple "
"--property=Restart=on-failure --property=RestartSec=1 "
"--property=TimeoutStartSec=30 "
"--property=StandardOutput=append:/tmp/lightscale-agent.log "
"--property=StandardError=append:/tmp/lightscale-agent.log -- "
+ cmd
)
node.wait_for_unit("lightscale-agent.service")
node.wait_until_succeeds("ip link show ls-natchurn", timeout=60)
start_agent(node1, ["10.0.0.1:51820"])
start_agent(node3, ["10.0.0.2:40000", "10.0.0.2:41000", "10.0.0.2:42000", "10.0.0.2:43000"])
data1 = json.loads(node1.succeed("cat /tmp/ls-state/state.json"))
data3 = json.loads(node3.succeed("cat /tmp/ls-state/state.json"))
node1_ip = data1["ipv4"]
node3_ip = data3["ipv4"]
def eventually_ping(src, dst, timeout=180):
src.wait_until_succeeds(
f"for i in $(seq 1 20); do ping -c 1 -W 1 {dst} && exit 0; sleep 1; done; exit 1",
timeout=timeout,
)
eventually_ping(node1, node3_ip)
eventually_ping(node3, node1_ip)
# Wait beyond short conntrack timeout and verify it re-establishes.
natgw.succeed("sleep 7")
natgw.execute("conntrack -L -p udp || true")
eventually_ping(node1, node3_ip)
eventually_ping(node3, node1_ip)
# Rebind multiple times.
for port in [41000, 42000]:
set_nat_port(port)
node1.wait_until_succeeds(f"wg show ls-natchurn endpoints | grep -q ':{port}'", timeout=180)
eventually_ping(node1, node3_ip)
eventually_ping(node3, node1_ip)
# Short full outage, then recover on a fresh mapped port.
natgw.succeed("iptables -I FORWARD -p udp --dport 51820 -j DROP")
natgw.succeed("iptables -I FORWARD -p udp --sport 51820 -j DROP")
node1.execute(f"ping -c 2 {node3_ip} || true")
natgw.succeed("sleep 10")
natgw.succeed("iptables -D FORWARD -p udp --dport 51820 -j DROP")
natgw.succeed("iptables -D FORWARD -p udp --sport 51820 -j DROP")
set_nat_port(43000)
node1.wait_until_succeeds("wg show ls-natchurn endpoints | grep -q ':43000'", timeout=240)
eventually_ping(node1, node3_ip, timeout=240)
eventually_ping(node3, node1_ip, timeout=240)
'';
}