lightscale/lab/test-soak.nix

{ pkgs, serverPkg, clientPkg }:
{
  name = "lightscale-lab-soak";
  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
        pkgs.iptables
      ];
    };
    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
        pkgs.iptables
      ];
    };
  };

  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 --setenv=LIGHTSCALE_ADMIN_TOKEN=test-admin -- "
        "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 'authorization: Bearer test-admin' "
        "-H 'content-type: application/json' "
        "-d '{\"name\":\"soak\",\"bootstrap_token_ttl_seconds\":600,"
        "\"bootstrap_token_uses\":10,\"bootstrap_token_tags\":[\"soak\"]}'"
    ))
    token = net["bootstrap_token"]["token"]

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

    enroll(node1, "node1", "10.0.0.1:51820")
    enroll(node2, "node2", "10.0.0.2:51820")

    def start_agent(node, endpoint):
        node.succeed("touch /tmp/lightscale-agent.log")
        cmd = (
            "lightscale-client --profile soak --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 1 "
            "--relay-reprobe-after 8 --stream-relay "
            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-soak", timeout=60)

    start_agent(node1, "10.0.0.1:51820")
    start_agent(node2, "10.0.0.2:51820")

    data1 = json.loads(node1.succeed("cat /tmp/ls-state/state.json"))
    data2 = json.loads(node2.succeed("cat /tmp/ls-state/state.json"))
    ip1 = data1["ipv4"]
    ip2 = data2["ipv4"]

    def mesh_ping(timeout=180):
        node1.wait_until_succeeds(f"ping -c 2 {ip2}", timeout=timeout)
        node2.wait_until_succeeds(f"ping -c 2 {ip1}", timeout=timeout)

    for cycle in range(1, 7):
        print(f"=== soak cycle {cycle} ===")
        mesh_ping()

        if cycle % 2 == 1:
            node1.succeed("iptables -I OUTPUT -p udp --dport 51820 -d 10.0.0.2 -j DROP")
            node1.succeed("iptables -I INPUT -p udp --sport 51820 -s 10.0.0.2 -j DROP")
            node2.succeed("iptables -I OUTPUT -p udp --dport 51820 -d 10.0.0.1 -j DROP")
            node2.succeed("iptables -I INPUT -p udp --sport 51820 -s 10.0.0.1 -j DROP")
            mesh_ping(timeout=240)
            node1.succeed("iptables -D OUTPUT -p udp --dport 51820 -d 10.0.0.2 -j DROP")
            node1.succeed("iptables -D INPUT -p udp --sport 51820 -s 10.0.0.2 -j DROP")
            node2.succeed("iptables -D OUTPUT -p udp --dport 51820 -d 10.0.0.1 -j DROP")
            node2.succeed("iptables -D INPUT -p udp --sport 51820 -s 10.0.0.1 -j DROP")
            node1.wait_until_succeeds("wg show ls-soak endpoints | grep -q '10.0.0.2:51820'", timeout=240)
            node2.wait_until_succeeds("wg show ls-soak endpoints | grep -q '10.0.0.1:51820'", timeout=240)
            mesh_ping(timeout=240)

        if cycle % 3 == 0:
            node2.succeed("systemctl stop lightscale-agent.service")
            start_agent(node2, "10.0.0.2:51820")
            mesh_ping(timeout=240)

    node1.succeed("! grep -qi panic /tmp/lightscale-agent.log")
    node2.succeed("! grep -qi panic /tmp/lightscale-agent.log")
  '';
}