Switch submodule remotes to GitHub mirrors
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

This commit is contained in:
centra 2026-02-24 02:41:01 +09:00
parent 169d207868
commit 559d6d2c6e
Signed by: centra
GPG key ID: 0C09689D20B25ACA
10 changed files with 1318 additions and 2 deletions

60
.github/workflows/linux-lab.yml vendored Normal file
View file

@ -0,0 +1,60 @@
name: linux-lab
on:
push:
branches:
- main
- master
pull_request:
workflow_dispatch:
permissions:
contents: read
jobs:
core:
name: core-${{ matrix.mode }}
runs-on: ubuntu-24.04
timeout-minutes: 180
strategy:
fail-fast: false
matrix:
mode: [fast, nat, netem]
steps:
- name: Checkout
uses: actions/checkout@v4
with:
submodules: recursive
- name: Install Nix
uses: cachix/install-nix-action@v31
with:
extra_nix_config: |
experimental-features = nix-command flakes
- name: Run core lab mode
run: ./lab/run.sh ${{ matrix.mode }}
extended:
if: github.event_name == 'workflow_dispatch'
name: extended-${{ matrix.mode }}
runs-on: ubuntu-24.04
timeout-minutes: 240
strategy:
fail-fast: false
matrix:
mode: [auth-url, nat-churn, relay-switch, standalone, soak]
steps:
- name: Checkout
uses: actions/checkout@v4
with:
submodules: recursive
- name: Install Nix
uses: cachix/install-nix-action@v31
with:
extra_nix_config: |
experimental-features = nix-command flakes
- name: Run extended lab mode
run: ./lab/run.sh ${{ matrix.mode }}

4
.gitmodules vendored
View file

@ -1,6 +1,6 @@
[submodule "lightscale-client"] [submodule "lightscale-client"]
path = lightscale-client path = lightscale-client
url = https://git.centraworks.net/centra/lightscale-client.git url = https://github.com/CentRa-Linux/lightscale-client.git
[submodule "lightscale-server"] [submodule "lightscale-server"]
path = lightscale-server path = lightscale-server
url = https://git.centraworks.net/centra/lightscale-server.git url = https://github.com/CentRa-Linux/lightscale-server.git

49
lab/run-linux-suite.sh Executable file
View file

@ -0,0 +1,49 @@
#!/usr/bin/env bash
set -euo pipefail
ROOT_DIR=$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)
cd "$ROOT_DIR"
SUITE="core"
WITH_ADMIN=0
for arg in "$@"; do
case "$arg" in
core|extended|all)
SUITE="$arg"
;;
--with-admin)
WITH_ADMIN=1
;;
*)
echo "unknown argument: $arg" >&2
echo "usage: ./lab/run-linux-suite.sh [core|extended|all] [--with-admin]" >&2
exit 1
;;
esac
done
CORE_MODES=(fast nat netem)
EXTENDED_MODES=(auth-url nat-churn relay-switch standalone soak)
case "$SUITE" in
core)
MODES=("${CORE_MODES[@]}")
;;
extended)
MODES=("${EXTENDED_MODES[@]}")
;;
all)
MODES=("${CORE_MODES[@]}" "${EXTENDED_MODES[@]}")
;;
esac
if [[ "$WITH_ADMIN" == "1" ]]; then
MODES+=(admin)
fi
for mode in "${MODES[@]}"; do
echo "=== running lab mode: $mode ==="
./lab/run.sh "$mode"
echo "=== finished lab mode: $mode ==="
done

142
lab/test-auth-url.nix Normal file
View file

@ -0,0 +1,142 @@
{ pkgs, serverPkg, clientPkg }:
{
name = "lightscale-lab-auth-url";
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 --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\":\"auth-url\",\"bootstrap_token_ttl_seconds\":600,"
"\"bootstrap_token_uses\":10,\"bootstrap_token_tags\":[\"auth-url\"]}'"
))
token = net["bootstrap_token"]["token"]
network_id = net["network"]["id"]
node1.succeed(
"lightscale-client --profile auth --config /tmp/ls-config.json "
"init http://10.0.0.1:8080"
)
node1.succeed(
f"lightscale-client --profile auth --config /tmp/ls-config.json "
f"--state-dir /tmp/ls-state register --node-name node1 -- {token}"
)
node2.succeed(
"lightscale-client --profile auth --config /tmp/ls-config.json "
"init http://10.0.0.1:8080"
)
node2.succeed(
"sh -c 'lightscale-client --profile auth --config /tmp/ls-config.json "
"--state-dir /tmp/ls-state register-url "
+ network_id +
" --node-name node2 --approve | tee /tmp/register-url.out'"
)
node2.succeed("grep -q 'approved=true' /tmp/register-url.out")
node2.succeed(
"lightscale-client --profile auth --config /tmp/ls-config.json "
"--state-dir /tmp/ls-state status | grep -q 'approved: true'"
)
node1.succeed(
"lightscale-client --profile auth --config /tmp/ls-config.json "
"--state-dir /tmp/ls-state heartbeat --endpoint 10.0.0.1:51820"
)
node2.succeed(
"lightscale-client --profile auth --config /tmp/ls-config.json "
"--state-dir /tmp/ls-state heartbeat --endpoint 10.0.0.2:51820"
)
def start_agent(node, endpoint):
node.succeed("touch /tmp/lightscale-agent.log")
cmd = (
"lightscale-client --profile auth --config /tmp/ls-config.json "
"--state-dir /tmp/ls-state agent --listen-port 51820 "
"--heartbeat-interval 5 --longpoll-timeout 5 "
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-auth", 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"]
node1.wait_until_succeeds(f"ping -c 3 {ip2}", timeout=180)
node2.wait_until_succeeds(f"ping -c 3 {ip1}", timeout=180)
'';
}

215
lab/test-nat-churn.nix Normal file
View file

@ -0,0 +1,215 @@
{ 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)
'';
}

202
lab/test-netem.nix Normal file
View file

@ -0,0 +1,202 @@
{ pkgs, serverPkg, clientPkg }:
{
name = "lightscale-lab-netem";
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 = "192.168.50.1"; 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 = "192.168.50.10"; prefixLength = 24; }
];
networking.defaultGateway = {
address = "192.168.50.1";
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 '192.168.50.1/24'")
node3.wait_until_succeeds("ip -4 addr show dev eth1 | grep -q '192.168.50.10/24'")
node3.wait_until_succeeds("ip -4 route show default | grep -q 'via 192.168.50.1'")
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 192.168.50.10:51820"
)
natgw.succeed(
f"iptables -t nat -A POSTROUTING -o eth1 -p udp -s 192.168.50.10 "
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)
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\":\"netem\",\"bootstrap_token_ttl_seconds\":600,"
"\"bootstrap_token_uses\":10,\"bootstrap_token_tags\":[\"netem\"]}'"
))
token = net["bootstrap_token"]["token"]
def enroll(node, name, endpoints):
node.succeed(
"lightscale-client --profile netem --config /tmp/ls-config.json "
"init http://10.0.0.1:8080"
)
node.succeed(
f"lightscale-client --profile netem --config /tmp/ls-config.json "
f"--state-dir /tmp/ls-state register --node-name {name} -- {token}"
)
cmd = (
"lightscale-client --profile netem --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"])
def start_agent(node, endpoints):
node.succeed("touch /tmp/lightscale-agent.log")
cmd = (
"lightscale-client --profile netem --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-netem", timeout=60)
start_agent(node1, ["10.0.0.1:51820"])
start_agent(node3, ["10.0.0.2:40000", "10.0.0.2:41000"])
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"]
node1.wait_until_succeeds(f"ping -c 3 {node3_ip}", timeout=180)
node3.wait_until_succeeds(f"ping -c 3 {node1_ip}", timeout=180)
# Impair WAN path between node1 and natgw to validate behavior under delay/loss.
natgw.succeed("tc qdisc replace dev eth1 root netem delay 120ms 20ms loss 8%")
natgw.succeed("tc -s qdisc show dev eth1 | grep -q netem")
def eventually_ping(src, dst):
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=180,
)
eventually_ping(node1, node3_ip)
eventually_ping(node3, node1_ip)
natgw.succeed("tc qdisc del dev eth1 root")
# Simulate NAT rebinding by changing external UDP port for node3.
set_nat_port(41000)
node1.wait_until_succeeds("wg show ls-netem endpoints | grep -q ':41000'", timeout=180)
node1.wait_until_succeeds(f"ping -c 3 {node3_ip}", timeout=180)
node3.wait_until_succeeds(f"ping -c 3 {node1_ip}", timeout=180)
'';
}

156
lab/test-relay-switch.nix Normal file
View file

@ -0,0 +1,156 @@
{ pkgs, serverPkg, clientPkg }:
{
name = "lightscale-lab-relay-switch";
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\":\"relay-switch\",\"bootstrap_token_ttl_seconds\":600,"
"\"bootstrap_token_uses\":10,\"bootstrap_token_tags\":[\"relay-switch\"]}'"
))
token = net["bootstrap_token"]["token"]
def enroll(node, name, endpoint):
node.succeed(
"lightscale-client --profile rly --config /tmp/ls-config.json "
"init http://10.0.0.1:8080"
)
node.succeed(
f"lightscale-client --profile rly --config /tmp/ls-config.json "
f"--state-dir /tmp/ls-state register --node-name {name} -- {token}"
)
node.succeed(
f"lightscale-client --profile rly --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")
node.execute("sh -c 'tail -n +1 -f /tmp/lightscale-agent.log >/dev/console 2>&1 &'")
cmd = (
"lightscale-client --profile rly --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-rly", 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"]
# Direct path baseline.
node1.wait_until_succeeds("wg show ls-rly endpoints | grep -q '10.0.0.2:51820'", timeout=120)
node2.wait_until_succeeds("wg show ls-rly endpoints | grep -q '10.0.0.1:51820'", timeout=120)
node1.wait_until_succeeds(f"ping -c 3 {ip2}", timeout=180)
node2.wait_until_succeeds(f"ping -c 3 {ip1}", timeout=180)
# Force direct UDP path down; stream relay should keep connectivity.
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")
node1.wait_until_succeeds("grep -q 'connected to 10.0.0.1:8443' /tmp/lightscale-agent.log", timeout=120)
node2.wait_until_succeeds("grep -q 'connected to 10.0.0.1:8443' /tmp/lightscale-agent.log", timeout=120)
node1.wait_until_succeeds(f"ping -c 3 {ip2}", timeout=240)
node2.wait_until_succeeds(f"ping -c 3 {ip1}", timeout=240)
# Bring UDP back and verify direct path recovers.
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-rly endpoints | grep -q '10.0.0.2:51820'", timeout=240)
node2.wait_until_succeeds("wg show ls-rly endpoints | grep -q '10.0.0.1:51820'", timeout=240)
node1.wait_until_succeeds(f"ping -c 3 {ip2}", timeout=240)
node2.wait_until_succeeds(f"ping -c 3 {ip1}", timeout=240)
'';
}

197
lab/test-simple-autoreg.nix Normal file
View file

@ -0,0 +1,197 @@
{ pkgs, serverPkg, clientPkg }:
let
serverModule = import ../nixos/modules/lightscale-server.nix {
defaultPackage = serverPkg;
};
clientModule = import ../nixos/modules/lightscale-client.nix {
defaultPackage = clientPkg;
};
in
{
name = "lightscale-lab-simple-autoreg";
nodes = {
server = { ... }: {
imports = [ serverModule ];
networking.hostName = "server";
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" ];
services.lightscale-server = {
enable = true;
listen = "10.0.0.1:8080";
stateFile = "/var/lib/lightscale-server/state.json";
adminToken = "lab-admin-token";
};
environment.systemPackages = [
clientPkg
pkgs.curl
pkgs.iputils
];
};
client1 = { ... }: {
imports = [ clientModule ];
networking.hostName = "client1";
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" ];
services.lightscale-client = {
enable = true;
controlUrls = [ "http://10.0.0.1:8080" ];
autoRegister = true;
enrollmentTokenFile = "/run/lightscale-enroll.token";
registerNodeName = "simple-client1";
registerExtraArgs = [
"--machine-private-key-file"
"/run/lightscale-machine.key"
"--wg-private-key-file"
"/run/lightscale-wg.key"
];
agentArgs = [
"--listen-port"
"51820"
"--apply-routes"
"--heartbeat-interval"
"5"
"--longpoll-timeout"
"5"
];
};
environment.systemPackages = [
clientPkg
pkgs.iputils
];
};
client2 = { ... }: {
imports = [ clientModule ];
networking.hostName = "client2";
networking.usePredictableInterfaceNames = false;
virtualisation.vlans = [ 1 ];
networking.interfaces.eth1.useDHCP = false;
networking.interfaces.eth1.ipv4.addresses = [
{
address = "10.0.0.3";
prefixLength = 24;
}
];
networking.firewall.enable = false;
boot.kernelModules = [ "wireguard" ];
services.lightscale-client = {
enable = true;
controlUrls = [ "http://10.0.0.1:8080" ];
autoRegister = true;
enrollmentTokenFile = "/run/lightscale-enroll.token";
registerNodeName = "simple-client2";
registerExtraArgs = [
"--machine-private-key-file"
"/run/lightscale-machine.key"
"--wg-private-key-file"
"/run/lightscale-wg.key"
];
agentArgs = [
"--listen-port"
"51820"
"--apply-routes"
"--heartbeat-interval"
"5"
"--longpoll-timeout"
"5"
];
};
environment.systemPackages = [
clientPkg
pkgs.iputils
];
};
};
testScript = ''
import base64
import json
start_all()
server.wait_for_unit("lightscale-server.service")
server.wait_for_open_port(8080, addr="10.0.0.1", timeout=120)
client1.wait_for_unit("multi-user.target")
client2.wait_for_unit("multi-user.target")
net = json.loads(server.succeed(
"curl -sSf -X POST http://10.0.0.1:8080/v1/networks "
"-H 'authorization: Bearer lab-admin-token' "
"-H 'content-type: application/json' "
"-d '{\"name\":\"simple-autoreg\",\"bootstrap_token_ttl_seconds\":1200,\"bootstrap_token_uses\":10}'"
))
token = net["bootstrap_token"]["token"]
network_id = net["network"]["id"]
def write_registration_files(node, machine_byte, wg_byte):
machine_key = base64.b64encode(bytes([machine_byte]) * 32).decode("ascii")
wg_key = base64.b64encode(bytes([wg_byte]) * 32).decode("ascii")
node.succeed(f"printf '%s\\n' '{token}' > /run/lightscale-enroll.token")
node.succeed(f"printf '%s\\n' '{machine_key}' > /run/lightscale-machine.key")
node.succeed(f"printf '%s\\n' '{wg_key}' > /run/lightscale-wg.key")
return machine_key, wg_key
def run_autoreg(node):
node.execute("systemctl reset-failed lightscale-client-register.service || true")
node.succeed("systemctl start lightscale-client-register.service")
node.wait_for_unit("lightscale-client-register.service")
node.wait_until_succeeds("test -s /var/lib/lightscale-client/default/state.json", timeout=120)
node.wait_until_succeeds("systemctl is-active lightscale-client.service", timeout=120)
node.wait_until_succeeds("ip link show ls-default", timeout=120)
expected1 = write_registration_files(client1, 11, 21)
expected2 = write_registration_files(client2, 12, 22)
run_autoreg(client1)
run_autoreg(client2)
state1 = json.loads(client1.succeed("cat /var/lib/lightscale-client/default/state.json"))
state2 = json.loads(client2.succeed("cat /var/lib/lightscale-client/default/state.json"))
assert state1["network_id"] == network_id
assert state2["network_id"] == network_id
assert state1["machine_private_key"] == expected1[0]
assert state1["wg_private_key"] == expected1[1]
assert state2["machine_private_key"] == expected2[0]
assert state2["wg_private_key"] == expected2[1]
ip1 = state1["ipv4"]
ip2 = state2["ipv4"]
server.wait_until_succeeds(
f"curl -sSf -H 'authorization: Bearer lab-admin-token' "
f"http://10.0.0.1:8080/v1/admin/networks/{network_id}/nodes | grep -q '\"approved\":true'",
timeout=120,
)
client1.wait_until_succeeds(f"ping -c 3 {ip2}", timeout=180)
client2.wait_until_succeeds(f"ping -c 3 {ip1}", timeout=180)
client1.succeed("systemctl restart lightscale-client.service")
client1.wait_for_unit("lightscale-client.service")
client1.wait_until_succeeds(f"ping -c 3 {ip2}", timeout=180)
'';
}

157
lab/test-soak.nix Normal file
View file

@ -0,0 +1,157 @@
{ 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")
'';
}

138
lab/test-standalone.nix Normal file
View file

@ -0,0 +1,138 @@
{ pkgs, serverPkg, clientPkg }:
{
name = "lightscale-lab-standalone";
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'")
# Start control plane as a plain process (no systemd-run usage).
node1.succeed("touch /tmp/lightscale-server.log")
node1.succeed(
"sh -c 'LIGHTSCALE_ADMIN_TOKEN=test-admin RUST_LOG=info "
"lightscale-server --listen 10.0.0.1:8080 --state /tmp/lightscale-state.json "
">/tmp/lightscale-server.log 2>&1 & echo $! >/tmp/lightscale-server.pid'"
)
node1.wait_until_succeeds("sh -c 'kill -0 $(cat /tmp/lightscale-server.pid)'", timeout=30)
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\":\"standalone\",\"bootstrap_token_ttl_seconds\":600,"
"\"bootstrap_token_uses\":10,\"bootstrap_token_tags\":[\"standalone\"]}'"
))
token = net["bootstrap_token"]["token"]
def enroll(node, name, endpoint):
node.succeed(
"lightscale-client --profile std --config /tmp/ls-config.json "
"init http://10.0.0.1:8080"
)
node.succeed(
f"lightscale-client --profile std --config /tmp/ls-config.json "
f"--state-dir /tmp/ls-state register --node-name {name} -- {token}"
)
node.succeed(
f"lightscale-client --profile std --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_daemon(node, endpoint):
node.succeed("touch /tmp/lightscale-daemon.log")
node.succeed(
"sh -c 'nohup lightscale-client --profile std --config /tmp/ls-config.json "
"--state-dir /tmp/ls-state daemon --profiles std "
"--agent-arg=--listen-port --agent-arg=51820 "
"--agent-arg=--heartbeat-interval --agent-arg=5 "
"--agent-arg=--longpoll-timeout --agent-arg=5 "
"--agent-arg=--endpoint --agent-arg=" + endpoint + " "
">/tmp/lightscale-daemon.log 2>&1 & echo $! >/tmp/lightscale-daemon.pid'"
)
node.wait_until_succeeds("sh -c 'kill -0 $(cat /tmp/lightscale-daemon.pid)'", timeout=30)
node.wait_until_succeeds("ip link show ls-std", timeout=120)
def stop_daemon(node):
node.execute("sh -c 'kill -TERM $(cat /tmp/lightscale-daemon.pid) || true'")
node.wait_until_succeeds(
"sh -c 'pid=$(cat /tmp/lightscale-daemon.pid); "
"if kill -0 \"$pid\" 2>/dev/null; then exit 1; else exit 0; fi'",
timeout=120,
)
start_daemon(node1, "10.0.0.1:51820")
start_daemon(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"]
node1.wait_until_succeeds(f"ping -c 3 {ip2}", timeout=180)
node2.wait_until_succeeds(f"ping -c 3 {ip1}", timeout=180)
# Restart plain daemons and verify connectivity recovers.
stop_daemon(node1)
stop_daemon(node2)
node1.fail("systemctl is-active lightscale-agent.service")
node2.fail("systemctl is-active lightscale-agent.service")
start_daemon(node1, "10.0.0.1:51820")
start_daemon(node2, "10.0.0.2:51820")
node1.wait_until_succeeds(f"ping -c 3 {ip2}", timeout=240)
node2.wait_until_succeeds(f"ping -c 3 {ip1}", timeout=240)
stop_daemon(node1)
stop_daemon(node2)
node1.execute("sh -c 'kill -TERM $(cat /tmp/lightscale-server.pid) || true'")
'';
}