lightscale/lab/test-userspace.nix

{ pkgs, serverPkg, clientPkg }:
{
  name = "lightscale-lab-userspace";
  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
      ];
    };
    node2 = { ... }: {
      networking.hostName = "node2";
      networking.usePredictableInterfaceNames = false;
      virtualisation.vlans = [ 1 ];
      networking.interfaces.eth1.useDHCP = false;
      networking.interfaces.eth1.ipv4.addresses = [
        { address = "10.0.0.2"; prefixLength = 24; }
      ];
      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")
    node2.wait_for_unit("multi-user.target")
    node1.wait_until_succeeds("ip -4 addr show dev eth1 | grep -q '10.0.0.1/24'")
    node2.wait_until_succeeds("ip -4 addr show dev eth1 | grep -q '10.0.0.2/24'")

    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 -- "
        "lightscale-server --listen 10.0.0.1:8080 --state /tmp/lightscale-state.json "
        "--stream-relay 10.0.0.1:8443 --stream-relay-listen 10.0.0.1:8443"
    )
    node1.wait_for_unit("lightscale-server.service")
    node1.wait_for_open_port(8080, addr="10.0.0.1", timeout=120)
    node1.wait_for_open_port(8443, 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 'content-type: application/json' "
        "-d '{\"name\":\"userspace\",\"bootstrap_token_ttl_seconds\":600," \
        "\"bootstrap_token_uses\":10,\"bootstrap_token_tags\":[\"userspace\"]}'"
    ))
    token = net["bootstrap_token"]["token"]

    def enroll(node, name, ip):
        node.succeed(
            "lightscale-client --profile userspace --config /tmp/ls-config.json "
            "init http://10.0.0.1:8080"
        )
        node.succeed(
            f"lightscale-client --profile userspace --config /tmp/ls-config.json "
            f"--state-dir /tmp/ls-state register --node-name {name} -- {token}"
        )
        node.succeed(
            f"lightscale-client --profile userspace --config /tmp/ls-config.json "
            f"--state-dir /tmp/ls-state heartbeat --endpoint {ip}:51820"
        )

    enroll(node1, "node1", "10.0.0.1")
    enroll(node2, "node2", "10.0.0.2")

    def start_agent(node, endpoints, stream_relay):
        node.succeed("touch /tmp/lightscale-agent.log")
        cmd = (
            "lightscale-client --profile userspace --config /tmp/ls-config.json "
            "--state-dir /tmp/ls-state agent --listen-port 51820 "
            "--heartbeat-interval 5 --longpoll-timeout 5 "
            "--backend boringtun --endpoint-stale-after 5 --endpoint-max-rotations 1"
        )
        for endpoint in endpoints:
            cmd += f" --endpoint {endpoint}"
        if stream_relay:
            cmd += " --stream-relay"
        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-userspace", timeout=60)

    # Phase 1: userspace endpoint rotation (invalid -> valid)
    start_agent(node1, ["203.0.113.1:51820", "10.0.0.1:51820"], False)
    start_agent(node2, ["203.0.113.2:51820", "10.0.0.2:51820"], False)

    data1 = json.loads(node1.succeed("cat /tmp/ls-state/state.json"))
    data2 = json.loads(node2.succeed("cat /tmp/ls-state/state.json"))
    nodes = [node1, node2]
    ips = [data1["ipv4"], data2["ipv4"]]

    def full_mesh_ping(nodes, ips):
        for i, src in enumerate(nodes):
            for j, dst in enumerate(nodes):
                if i == j:
                    continue
                src.wait_until_succeeds(f"ping -c 3 {ips[j]}", timeout=120)

    full_mesh_ping(nodes, ips)

    # Phase 2: userspace stream relay fallback
    node1.succeed("systemctl stop lightscale-agent.service")
    node2.succeed("systemctl stop lightscale-agent.service")
    start_agent(node1, ["203.0.113.1:51820"], True)
    start_agent(node2, ["203.0.113.2:51820"], True)

    full_mesh_ping(nodes, ips)
  '';
}