scripts: consolidate Agent deployment and remove Probe packages

Use Agent for native, container, systemd and OpenWrt installations and signed updates. Remove retired Probe deployment templates and smoke scripts.

scripts/package.sh

@@ -1,12 +1,13 @@
 #!/usr/bin/env bash
-# 打包：scripts/package.sh <agent|probe> <rid> → backend/artifacts/packages/moshiro-<role>-<version>-<rid>.tar.gz（附 .sha256）
+# 打包：scripts/package.sh <agent> <rid> → backend/artifacts/packages/moshiro-<role>-<version>-<rid>.tar.gz（附 .sha256）
 # 内容：节点二进制（agent 还带 fleet-enroll）、systemd 单元、OpenWrt procd 脚本、install.sh。
 # 发布产物要先用 scripts/publish.sh 生成（agent 需要 agent 和 fleet-enroll 两个）。
 set -euo pipefail
-role="${1:?usage: package.sh <agent|probe> <rid>}"
-rid="${2:?usage: package.sh <agent|probe> <rid>}"
+role="${1:?usage: package.sh <agent> <rid>}"
+rid="${2:?usage: package.sh <agent> <rid>}"
+[[ "$role" == agent ]] || { echo "only agent packages are supported" >&2; exit 2; }
 root="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
-version="$(sed -n 's#.*<Version>\(.*\)</Version>.*#\1#p' "$root/Directory.Build.props" | head -1)"
+version="$(sed -n 's#.*<Version>\(.*\)</Version>.*#\1#p' "$root/backend/Directory.Build.props" | head -1)"
 name="moshiro-$role-$version-$rid"
 stage="$root/backend/artifacts/packages/stage/$name"
 quic_runtime="$root/backend/artifacts/packages/quic-runtime/$rid"
@@ -27,10 +28,11 @@ fi
 cp "$root/backend/artifacts/publish/$role/$rid/moshiro-$role" "$stage/"
 cp "$root/packaging/install.sh" "$stage/"
 cp "$root/packaging/systemd/moshiro-$role.service" "$stage/systemd/"
-cp "$root/$role/README.md" "$stage/README.md"
+cp "$root/backend/$role/README.md" "$stage/README.md"
 if [ "$role" = agent ]; then
   cp "$root/backend/artifacts/publish/fleet-enroll/$rid/fleet-enroll" "$stage/"
-else
+fi
+if [[ "$rid" == linux-* ]]; then
   mkdir -p "$stage/openwrt"
   cp "$root/packaging/openwrt/"* "$stage/openwrt/"
 fi

scripts/publish.sh

@@ -1,16 +1,15 @@
 #!/usr/bin/env bash
-# 发布一个节点程序：scripts/publish.sh <agent|probe|backend|fleet-enroll> <rid>
+# 发布一个节点程序：scripts/publish.sh <agent|backend|fleet-enroll> <rid>
 # 输出到 backend/artifacts/publish/<project>/<rid>/。linux-musl-x64（Alpine）和 linux-x64（Ubuntu noble，glibc）在 NativeAOT SDK 容器里构建（任何主机都可以，包括 Windows），
 # 其他 rid 用本机的 dotnet（NativeAOT 不能交叉编译到别的操作系统，所以 win-x64 要在 Windows 上、linux-x64 要在 glibc 的 Linux 上构建）。
 set -euo pipefail
 
-project="${1:?usage: publish.sh <agent|probe|backend|fleet-enroll> <rid>}"
-rid="${2:?usage: publish.sh <agent|probe|backend|fleet-enroll> <rid>}"
+project="${1:?usage: publish.sh <agent|backend|fleet-enroll> <rid>}"
+rid="${2:?usage: publish.sh <agent|backend|fleet-enroll> <rid>}"
 root="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
 
 case "$project" in
   agent) csproj="backend/agent/src/MoShiro.Agent" ;;
-  probe) csproj="backend/probe/src/MoShiro.Probe" ;;
   backend) csproj="backend/src/MoShiro.Backend" ;;
   fleet-enroll) csproj="backend/fleet/tools/fleet-enroll" ;;
   *) echo "unknown project: $project" >&2; exit 2 ;;

scripts/smoke-linux.sh

@@ -1,86 +0,0 @@
-#!/bin/sh
-# 在 Linux 容器里用发布出来的 NativeAOT 二进制跑一遍完整链路：后端 → 注册 Agent → Agent → Probe → 数据落库。
-#
-#   RID=linux-musl-x64 IMAGE=alpine:3.22          （默认）
-#   RID=linux-x64      IMAGE=debian:bookworm-slim
-#   MODE=install       用打包产物里的 install.sh 安装（agent 注册、probe 身份、以 moshiro 用户运行），而不是直接运行二进制
-#
-# 用法（在仓库根目录；Windows 下先 export MSYS_NO_PATHCONV=1；需先运行 scripts/package.sh）：
-#   docker run --rm -e RID -e MODE -v "$PWD/backend/artifacts:/art:ro" -v "$PWD/scripts/smoke-linux.sh:/smoke.sh:ro" $IMAGE sh /smoke.sh [秒数]
-set -eu
-seconds="${1:-60}"
-rid="${RID:-linux-musl-x64}"
-mode="${MODE:-direct}"
-pub=/art/publish
-
-if command -v apk >/dev/null 2>&1; then
-  apk add --no-cache curl jq libssl3 libstdc++ >/dev/null
-else
-  apt-get update >/dev/null && apt-get install -y --no-install-recommends curl jq ca-certificates libssl3 >/dev/null
-fi
-
-work=/tmp/smoke
-mkdir -p "$work/data" "$work/probe"
-base=http://127.0.0.1:18080
-
-export ASPNETCORE_URLS="$base" MoShiro__DataDirectory="$work/data" MoShiro__InitialAdminUser=admin \
-  MoShiro__InitialAdminPassword="correct horse battery" MoShiro__SecureCookies=false
-(cd "$pub/backend/$rid" && ./moshiro-backend >"$work/backend.log" 2>&1 &)
-for i in $(seq 1 60); do curl -fs "$base/healthz" >/dev/null 2>&1 && break; sleep 1; done
-curl -fs "$base/healthz" >/dev/null || { cat "$work/backend.log"; echo "backend did not start"; exit 1; }
-echo "backend up"
-
-login=$(curl -si -X POST "$base/api/admin/auth/login" -H 'Content-Type: application/json' -d '{"username":"admin","password":"correct horse battery"}')
-cookie=$(echo "$login" | grep -i '^set-cookie' | head -1 | sed 's/^[^:]*: //; s/;.*//' | tr -d '\r')
-csrf=$(echo "$login" | tail -1 | jq -r .csrf_token)
-token=$(curl -fs -X POST "$base/api/admin/api-tokens" -H "Cookie: $cookie" -H "X-CSRF-Token: $csrf" -H 'Content-Type: application/json' -d '{"name":"smoke","scope":"admin"}' | jq -r .plaintext)
-auth="Authorization: Bearer $token"
-join=$(curl -fs -X POST "$base/api/admin/join-tokens" -H "$auth" -H 'Content-Type: application/json' -d '{"auto_approve":true,"max_uses":1,"expires_in_hours":1}' | jq -r .plaintext)
-probe_join=$(curl -fs -X POST "$base/api/admin/join-tokens" -H "$auth" -H 'Content-Type: application/json' -d '{"role":"probe","max_uses":1,"expires_in_hours":1}' | jq -r .plaintext)
-
-if [ "$mode" = install ]; then
-  # 安装到 /usr/lib/moshiro 和 /etc/moshiro，以 moshiro 用户运行（没有 systemd：自己启动）
-  mkdir -p /tmp/pkg && for role in agent probe; do tar -C /tmp/pkg -xzf /art/packages/moshiro-$role-*-$rid.tar.gz; done
-  (cd /tmp/pkg/moshiro-agent-* && ./install.sh agent --server "$base" --join "$join" --name smoke-agent --no-start)
-  ls -l /usr/lib/moshiro /etc/moshiro
-  run_as() { su -s /bin/sh moshiro -c "exec $*"; }
-  run_as /usr/lib/moshiro/moshiro-agent --config /etc/moshiro/agent.json >"$work/agent.log" 2>&1 &
-  agent_pid=$!
-  sleep 3
-  (cd /tmp/pkg/moshiro-probe-* && ./install.sh probe --server "$base" --join "$probe_join" --name smoke-probe --no-start)
-  run_as /usr/lib/moshiro/moshiro-probe --config /etc/moshiro/probe.json >"$work/probe.log" 2>&1 &
-  probe_pid=$!
-else
-  for role in agent probe; do
-    package_dir="$work/$role-bin"
-    mkdir -p "$package_dir"
-    set -- /art/packages/moshiro-"$role"-*-"$rid".tar.gz
-    [ -f "$1" ] || { echo "missing QUIC-bundled package: $1 (run scripts/package.sh first)" >&2; exit 1; }
-    tar -xzf "$1" -C "$package_dir"
-  done
-  mkdir -p "$work/agent-id"
-  "$work/agent-bin/fleet-enroll" --server "$base" --join "$join" --name smoke-agent --identity-dir "$work/agent-id"
-  printf '{"fleet":{"server":"%s","identity_dir":"%s"},"listen":{"bind":"127.0.0.1"}}' "$base" "$work/agent-id" >"$work/agent.json"
-  "$work/agent-bin/moshiro-agent" --config "$work/agent.json" >"$work/agent.log" 2>&1 &
-  agent_pid=$!
-  sleep 3
-  mkdir -p "$work/probe-id"
-  "$work/probe-bin/fleet-enroll" --server "$base" --join "$probe_join" --name smoke-probe --identity-dir "$work/probe-id"
-  printf '{"fleet":{"server":"%s","identity_dir":"%s"}}' "$base" "$work/probe-id" >"$work/probe/probe.json"
-  "$work/probe-bin/moshiro-probe" --config "$work/probe/probe.json" >"$work/probe.log" 2>&1 &
-  probe_pid=$!
-fi
-
-rss() { awk '/VmRSS/ {printf "%.1f", $2/1024}' "/proc/$1/status"; }
-i=0
-while [ "$i" -lt "$seconds" ]; do
-  sleep 10; i=$((i + 10))
-  echo "t+${i}s probe=$(rss "$probe_pid")MB agent=$(rss "$agent_pid")MB"
-done
-
-echo "agents: $(curl -fs "$base/api/admin/agents" -H "$auth" | jq -c '[.items[] | {name, status, probe_connected}]')"
-echo "chunks: $(curl -fs "$base/api/admin/raw/chunks" -H "$auth" | jq -c '[.items[].kind] | group_by(.) | map({(.[0]): length}) | add')"
-agent_id=$(curl -fs "$base/api/admin/agents" -H "$auth" | jq -r '.items[0].id')
-echo "tcp summary: $(curl -fs "$base/api/public/lines/$agent_id/series?range=5m&step=1s&metrics=tcp.rtt.p50,tcp.rtt.p999,tcp.rtt.max,tcp.rtt.tail,tcp.slow,tcp.retrans,tcp.srtt.p95" -H "$auth" | jq -c '.summary.metrics')"
-echo "--- probe.log"; tail -4 "$work/probe.log"
-echo "--- agent.log"; tail -4 "$work/agent.log"

scripts/smoke-systemd.sh

@@ -1,31 +0,0 @@
-# 在带 systemd 的 Debian 容器里用 install.sh 安装并以 systemd 单元运行（验证沙箱指令与 NativeAOT 兼容）。
-# docker run -d --name sysd --privileged --cgroupns=host -v /sys/fs/cgroup:/sys/fs/cgroup:rw -v "$PWD/backend/artifacts:/art:ro" jrei/systemd-debian:12
-# docker cp scripts/smoke-systemd.sh sysd:/s.sh && docker exec sysd bash /s.sh
-set -eu
-apt-get update >/dev/null && apt-get install -y --no-install-recommends curl jq ca-certificates libssl3 >/dev/null
-base=http://127.0.0.1:18080
-mkdir -p /tmp/data
-cd /art/publish/backend/linux-x64
-ASPNETCORE_URLS=$base MoShiro__DataDirectory=/tmp/data MoShiro__InitialAdminUser=admin MoShiro__InitialAdminPassword="correct horse battery" MoShiro__SecureCookies=false nohup ./moshiro-backend >/tmp/backend.log 2>&1 &
-for i in $(seq 1 60); do curl -fs $base/healthz >/dev/null 2>&1 && break; sleep 1; done
-login=$(curl -si -X POST $base/api/admin/auth/login -H 'Content-Type: application/json' -d '{"username":"admin","password":"correct horse battery"}')
-cookie=$(echo "$login" | grep -i '^set-cookie' | head -1 | sed 's/^[^:]*: //; s/;.*//' | tr -d '\r')
-csrf=$(echo "$login" | tail -1 | jq -r .csrf_token)
-token=$(curl -fs -X POST $base/api/admin/api-tokens -H "Cookie: $cookie" -H "X-CSRF-Token: $csrf" -H 'Content-Type: application/json' -d '{"name":"t","scope":"admin"}' | jq -r .plaintext)
-auth="Authorization: Bearer $token"
-join=$(curl -fs -X POST $base/api/admin/join-tokens -H "$auth" -H 'Content-Type: application/json' -d '{"auto_approve":true,"max_uses":1,"expires_in_hours":1}' | jq -r .plaintext)
-mkdir -p /tmp/pkg; for r in agent probe; do tar -C /tmp/pkg -xzf /art/packages/moshiro-$r-*-linux-x64.tar.gz; done
-(cd /tmp/pkg/moshiro-agent-* && ./install.sh agent --server $base --join "$join" --name sysd-agent)
-sleep 3
-probe_join=$(curl -fs -X POST $base/api/admin/join-tokens -H "$auth" -H 'Content-Type: application/json' -d '{"role":"probe","max_uses":1,"expires_in_hours":1}' | jq -r .plaintext)
-(cd /tmp/pkg/moshiro-probe-* && ./install.sh probe --server $base --join "$probe_join" --name sysd-probe)
-sleep 25
-systemctl is-active moshiro-agent moshiro-probe
-echo "agents: $(curl -fs $base/api/admin/agents -H "$auth" | jq -c '[.items[] | {name, status, probe_connected}]')"
-echo "chunks: $(curl -fs $base/api/admin/raw/chunks -H "$auth" | jq -c '[.items[].kind] | length')"
-systemctl show moshiro-agent -p MemoryCurrent -p NRestarts | tr '\n' ' '; systemctl show moshiro-probe -p MemoryCurrent -p NRestarts | tr '\n' ' '; echo
-ps -o user,rss,comm -C moshiro-agent,moshiro-probe
-journalctl -u moshiro-agent -u moshiro-probe --no-pager | tail -8
-echo "--- sandbox check: agent cannot write /etc/moshiro"; nsenter -t $(systemctl show moshiro-agent -p MainPID --value) -m sh -c 'touch /etc/moshiro/x 2>&1; ls /proc/self/status >/dev/null' 2>&1 | head -2
-grep -E "NoNewPrivs|CapEff|Seccomp:" /proc/$(systemctl show moshiro-agent -p MainPID --value)/status
-grep -E "NoNewPrivs|CapEff|Seccomp:" /proc/$(systemctl show moshiro-probe -p MainPID --value)/status

scripts/smoke-wsl.py

@@ -1,77 +0,0 @@
-"""跨机器的真实链路验证：Windows 上的 Probe（NativeAOT exe）→ WSL2 里的 Agent 和 Backend（真实的虚拟网络、不同的单调时钟原点）。
-
-前提：WSL 里有 Debian（root）、tc（netem 注入丢包/延迟/黑洞用）；已经用 scripts/publish.sh 发布了 linux-x64 的 agent、backend、fleet-enroll，
-并且 Windows 上发布了 probe（dotnet publish backend/probe/src/MoShiro.Probe -c Release -r win-x64 -p:PublishAot=true -o backend/artifacts/aot-probe）。
-把 linux-x64 的产物复制到 WSL 的 /root/mon/，在 WSL 里用 setsid -f nohup 启动 moshiro-backend（监听 0.0.0.0:29181），再运行本脚本。
-故障注入见 docs/ops/deployment.md 第 3 节。
-"""
-import json, os, subprocess, sys, time, urllib.request, urllib.error
-
-# 要连的是本机的虚拟网络（WSL），不能走环境变量里的 HTTP(S)_PROXY
-OPENER = urllib.request.build_opener(urllib.request.ProxyHandler({}))
-
-ROOT = r"C:\Users\JustMe\Desktop\moshiro-proj"
-PROBE_EXE = os.path.join(ROOT, "backend", "artifacts", "aot-probe", "moshiro-probe.exe")
-WORK = r"C:\Users\JustMe\AppData\Local\Temp\realnet"
-PORT = 29181
-
-
-def wsl(cmd, check=True):
-    r = subprocess.run(["wsl", "-d", "Debian", "-u", "root", "-e", "bash", "-c", cmd], capture_output=True, text=True)
-    if check and r.returncode != 0:
-        print("WSL FAILED:", cmd, r.stdout, r.stderr)
-        raise SystemExit(1)
-    return r.stdout.strip()
-
-
-WSLIP = wsl("ip -4 addr show eth0 | grep -oP '(?<=inet )[0-9.]+'")
-BASE = f"http://{WSLIP}:{PORT}"
-
-
-def req(method, path, body=None, headers=None, ctype="application/json"):
-    data = None if body is None else json.dumps(body).encode()
-    h = {"Content-Type": ctype}
-    h.update(headers or {})
-    r = urllib.request.Request(BASE + path, data=data, method=method, headers=h)
-    try:
-        with OPENER.open(r, timeout=30) as resp:
-            text = resp.read().decode()
-            return resp, (json.loads(text) if text else None)
-    except urllib.error.HTTPError as e:
-        print("HTTP", e.code, e.read().decode())
-        raise
-
-
-def main():
-    os.makedirs(WORK, exist_ok=True)
-    print("wsl ip", WSLIP)
-    resp, login = req("POST", "/api/admin/auth/login", {"username": "admin", "password": os.environ.get("MOSHIRO_ADMIN_PASSWORD", "correct horse battery")})
-    cookie = resp.headers["Set-Cookie"].split(";")[0]
-    _, tok = req("POST", "/api/admin/api-tokens", {"name": "net", "scope": "admin"}, {"Cookie": cookie, "X-CSRF-Token": login["csrf_token"]})
-    auth = {"Authorization": "Bearer " + tok["plaintext"]}
-    json.dump({"base": BASE, "token": tok["plaintext"], "wslip": WSLIP}, open(WORK + r"\state.json", "w"))
-    _, join = req("POST", "/api/admin/join-tokens", {"auto_approve": True, "max_uses": 1, "expires_in_hours": 1}, auth)
-
-    wsl("pkill moshiro-agent; cd /root/mon && rm -rf id agent.json")
-    print(wsl(f"cd /root/mon && ./fleet-enroll --server http://127.0.0.1:{PORT} --join {join['plaintext']} --name wsl-agent --identity-dir /root/mon/id"))
-    cfg = {"fleet": {"server": f"http://127.0.0.1:{PORT}", "identity_dir": "/root/mon/id"}, "listen": {"bind": "0.0.0.0"}}
-    wsl("cat > /root/mon/agent.json <<'EOF'\n" + json.dumps(cfg) + "\nEOF")
-    wsl("cd /root/mon && (setsid nohup ./moshiro-agent --config /root/mon/agent.json > agent.log 2>&1 < /dev/null &)")
-    time.sleep(3)
-    _, agents = req("GET", "/api/admin/agents", None, auth)
-    agent = agents["items"][0]
-    print("agent", agent["id"], agent["status"], "online", agent.get("online"))
-    # the backend only sees 127.0.0.1 as the agent's address: the panel needs the address the probe can reach
-    req("PATCH", "/api/admin/agents/" + agent["id"], {"public_host": WSLIP}, auth, "application/merge-patch+json")
-    _, ident = req("POST", "/api/admin/probes", {"name": "home"}, auth)
-    print("bootstrap", ident["bootstrap"])
-    os.makedirs(WORK + r"\probe", exist_ok=True)
-    json.dump(ident, open(WORK + r"\probe\probe.json", "w"))
-    json.dump({"base": BASE, "token": tok["plaintext"], "wslip": WSLIP, "agent_id": agent["id"]}, open(WORK + r"\state.json", "w"))
-    log = open(WORK + r"\probe.log", "w")
-    p = subprocess.Popen([PROBE_EXE, "--config", WORK + r"\probe\probe.json"], stdout=log, stderr=subprocess.STDOUT)
-    open(WORK + r"\probe.pid", "w").write(str(p.pid))
-    print("probe pid", p.pid)
-
-
-main()

scripts/stability_wsl.py

@@ -1,613 +0,0 @@
-"""
-稳定性实测：真实的初始化、正常关闭、崩溃关闭、链路故障与恢复。
-
-拓扑：Windows 上的 Probe（NativeAOT exe）→ WSL2 里的 Agent → 同一个 WSL 里的 Backend。
-Probe↔Agent 走真实的虚拟网络（eth0），Agent↔Backend 走 WSL 里的回环（故障用 tc 注入在 lo 上）。
-
-前提：WSL 里有 Debian（root）和 tc；已发布 linux-x64 的 agent、backend、fleet-enroll（scripts/publish.sh），
-以及 Windows 的 probe（dotnet publish backend/probe/src/MoShiro.Probe -c Release -r win-x64 -p:PublishAot=true -o backend/artifacts/aot-probe）。
-
-用法：python scripts/stability_wsl.py [场景名 ...]   （不带参数运行全部）
-场景：init  graceful  crash  probe_link  backend_link  silent  soak
-"""
-import ctypes
-import json
-import os
-import subprocess
-import sys
-import time
-import urllib.error
-import urllib.request
-
-# 要连的是本机的虚拟网络（WSL），不能走环境变量里的 HTTP(S)_PROXY
-OPENER = urllib.request.build_opener(urllib.request.ProxyHandler({}))
-
-ROOT = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
-PROBE_EXE = os.path.join(ROOT, "backend", "artifacts", "aot-probe", "moshiro-probe.exe")
-PUB = "/mnt/" + ROOT[0].lower() + ROOT[2:].replace("\\", "/") + "/backend/artifacts/publish"
-WORK = os.path.join(os.environ.get("TEMP", "."), "stability")
-PORT = 29181
-ADMIN_PASSWORD = os.environ.get("MOSHIRO_ADMIN_PASSWORD", "correct horse battery")  # 测试用后端的初始管理员密码
-RESULTS = []
-HIDDEN = subprocess.CREATE_NO_WINDOW  # 所有子进程都不弹窗
-
-
-# ------------------------------------------------------------------ WSL / process helpers
-
-def wsl(cmd, check=True, timeout=120):
-    r = subprocess.run(["wsl", "-d", "Debian", "-u", "root", "-e", "bash", "-c", cmd], capture_output=True, text=True, timeout=timeout, creationflags=HIDDEN)
-    if check and r.returncode != 0:
-        raise RuntimeError(f"wsl failed ({r.returncode}): {cmd}\n{r.stdout}\n{r.stderr}")
-    return r.stdout.strip()
-
-
-WSLIP = wsl("ip -4 addr show eth0 | grep -oP '(?<=inet )[0-9.]+'")
-BASE = f"http://{WSLIP}:{PORT}"
-TOKEN = ""
-
-LAUNCHER = r"""#!/bin/sh
-# start.sh <name> <command...>: run detached, append to <name>.log, record the exit code in <name>.exit
-name=$1; shift
-rm -f /root/mon/$name.exit
-setsid -f sh -c 'cd /root/mon; "$@" >> /root/mon/'$name'.log 2>&1; echo $? > /root/mon/'$name'.exit' sh "$@" < /dev/null
-"""
-
-
-def now():
-    return time.time()
-
-
-def check(name, ok, detail=""):
-    RESULTS.append((name, bool(ok), detail))
-    print(f"  [{'PASS' if ok else 'FAIL'}] {name}" + (f" - {detail}" if detail else ""), flush=True)
-    return ok
-
-
-def wait_until(cond, timeout, step=0.5):
-    start = now()
-    while now() - start < timeout:
-        try:
-            if cond():
-                return now() - start
-        except Exception:
-            pass
-        time.sleep(step)
-    return None
-
-
-def api(method, path, body=None, ctype="application/json", timeout=15):
-    headers = {"Content-Type": ctype}
-    if TOKEN:
-        headers["Authorization"] = "Bearer " + TOKEN
-    data = None if body is None else json.dumps(body).encode()
-    req = urllib.request.Request(BASE + path, data=data, method=method, headers=headers)
-    with OPENER.open(req, timeout=timeout) as resp:
-        text = resp.read().decode()
-        return json.loads(text) if text else None
-
-
-def backend_up():
-    try:
-        OPENER.open(BASE + "/healthz", timeout=2)
-        return True
-    except Exception:
-        return False
-
-
-# ------------------------------------------------------------------ components
-
-def pids(name):
-    out = wsl(f"pgrep -x {name} || true")
-    return [int(x) for x in out.split()]
-
-
-def rss_mb(name):
-    p = pids(name)
-    if not p:
-        return None
-    return int(wsl(f"ps -o rss= -p {p[0]}")) / 1024
-
-
-def start_backend():
-    wsl(f"cd /root/mon && /root/mon/start.sh backend env ASPNETCORE_URLS=http://0.0.0.0:{PORT} MoShiro__DataDirectory=/root/mon/data "
-        f"MoShiro__InitialAdminUser=admin 'MoShiro__InitialAdminPassword={ADMIN_PASSWORD}' MoShiro__SecureCookies=false ./moshiro-backend")
-    return wait_until(backend_up, 30)
-
-
-def start_agent():
-    wsl("/root/mon/start.sh agent ./moshiro-agent --config /root/mon/agent.json")
-    return wait_until(lambda: bool(pids("moshiro-agent")), 10)
-
-
-def stop_proc(name, sig="TERM", timeout=30):
-    """Send a signal and wait for the process to exit; returns (seconds, exit code or None)."""
-    start = now()
-    wsl(f"pkill -{sig} -x {name} || true")
-    gone = wait_until(lambda: not pids(name), timeout, 0.2)
-    code = wsl(f"cat /root/mon/{name.replace('moshiro-', '')}.exit 2>/dev/null || true")
-    return (now() - start if gone is not None else None), (int(code) if code.strip().lstrip('-').isdigit() else None)
-
-
-def send_ctrl_break(pid):
-    """
-    Ctrl+Break to a process that owns its own (hidden) console: attach to that console, broadcast the event, detach.
-    Ctrl+Break rather than Ctrl+C: a console process can inherit "ignore Ctrl+C" from its parent (it does under most job runners),
-    Ctrl+Break cannot be ignored that way. .NET reports it as SIGQUIT.
-    """
-    from ctypes import wintypes
-    k = ctypes.WinDLL("kernel32", use_last_error=True)
-    handler = ctypes.WINFUNCTYPE(wintypes.BOOL, wintypes.DWORD)(lambda event: True)  # swallow the event in this process
-    k.FreeConsole()
-    if not k.AttachConsole(pid):
-        raise OSError(f"AttachConsole({pid}) failed: {ctypes.get_last_error()}")
-    k.SetConsoleCtrlHandler(handler, True)
-    k.GenerateConsoleCtrlEvent(1, 0)
-    time.sleep(0.3)
-    k.FreeConsole()
-    k.SetConsoleCtrlHandler(handler, False)
-
-
-class Probe:
-    def __init__(self):
-        self.proc = None
-        self.log = os.path.join(WORK, "probe.log")
-
-    def start(self):
-        os.makedirs(WORK, exist_ok=True)
-        handle = open(self.log, "a")
-        # 自己的隐藏控制台：这样才能从外面给它发 Ctrl+Break
-        flags = subprocess.CREATE_NO_WINDOW
-        self.proc = subprocess.Popen([PROBE_EXE, "--config", os.path.join(WORK, "probe", "probe.json")], stdout=handle, stderr=subprocess.STDOUT, creationflags=flags)
-        return self.proc.pid
-
-    def alive(self):
-        return self.proc is not None and self.proc.poll() is None
-
-    def stop(self, mode="graceful", timeout=20):
-        if not self.alive():
-            return 0.0, self.proc.poll() if self.proc else None
-        start = now()
-        if mode == "graceful":
-            send_ctrl_break(self.proc.pid)
-        else:
-            self.proc.kill()
-        try:
-            code = self.proc.wait(timeout)
-        except subprocess.TimeoutExpired:
-            self.proc.kill()
-            return None, None
-        return now() - start, code
-
-    def suspend(self):
-        """Freeze the process (its TCP connections stay open and go silent, like a hung machine)."""
-        handle = ctypes.windll.kernel32.OpenProcess(0x0800, False, self.proc.pid)  # PROCESS_SUSPEND_RESUME
-        ctypes.windll.ntdll.NtSuspendProcess(handle)
-        ctypes.windll.kernel32.CloseHandle(handle)
-
-    def resume(self):
-        handle = ctypes.windll.kernel32.OpenProcess(0x0800, False, self.proc.pid)
-        ctypes.windll.ntdll.NtResumeProcess(handle)
-        ctypes.windll.kernel32.CloseHandle(handle)
-
-    def working_set_mb(self):
-        if not self.alive():
-            return None
-        out = subprocess.run(["powershell", "-NoProfile", "-Command", f"(Get-Process -Id {self.proc.pid}).WorkingSet64"], capture_output=True, text=True, creationflags=HIDDEN).stdout.strip()
-        return int(out) / 1048576 if out else None
-
-    def log_text(self):
-        return open(self.log, encoding="utf-8", errors="replace").read() if os.path.exists(self.log) else ""
-
-
-PROBE = Probe()
-AGENT_ID = ""
-
-
-# ------------------------------------------------------------------ environment setup
-
-def setup(fresh=True):
-    """Clean state, copy the artifacts, start backend/agent, create the probe identity. The probe is not started."""
-    global TOKEN, AGENT_ID
-    PROBE.stop("kill")
-    wsl("pkill -9 -x moshiro-agent; pkill -9 -x moshiro-backend; tc qdisc del dev eth0 ingress 2>/dev/null; tc qdisc del dev eth0 root 2>/dev/null; tc qdisc del dev lo root 2>/dev/null; tc qdisc del dev lo ingress 2>/dev/null; true")
-    if fresh:
-        wsl(f"rm -rf /root/mon && mkdir -p /root/mon && cp {PUB}/backend/linux-x64/* {PUB}/agent/linux-x64/* {PUB}/fleet-enroll/linux-x64/* /root/mon/ && chmod +x /root/mon/moshiro-* /root/mon/fleet-enroll")
-        wsl("cat > /root/mon/start.sh <<'EOF'\n" + LAUNCHER + "EOF\nchmod +x /root/mon/start.sh")
-        if os.path.isdir(WORK):
-            subprocess.run(["cmd", "/c", "rmdir", "/s", "/q", WORK], capture_output=True, creationflags=HIDDEN)
-        os.makedirs(os.path.join(WORK, "probe"), exist_ok=True)
-    assert start_backend() is not None, "backend did not start"
-    if fresh:
-        req = urllib.request.Request(BASE + "/api/admin/auth/login", data=json.dumps({"username": "admin", "password": ADMIN_PASSWORD}).encode(), headers={"Content-Type": "application/json"}, method="POST")
-        with OPENER.open(req, timeout=15) as resp:
-            cookie = resp.headers["Set-Cookie"].split(";")[0]
-            csrf = json.loads(resp.read())["csrf_token"]
-        req = urllib.request.Request(BASE + "/api/admin/api-tokens", data=json.dumps({"name": "stab", "scope": "admin"}).encode(), headers={"Content-Type": "application/json", "Cookie": cookie, "X-CSRF-Token": csrf}, method="POST")
-        TOKEN = json.loads(OPENER.open(req, timeout=15).read())["plaintext"]
-        join = api("POST", "/api/admin/join-tokens", {"auto_approve": True, "max_uses": 5, "expires_in_hours": 1})["plaintext"]
-        wsl(f"cd /root/mon && ./fleet-enroll --server http://127.0.0.1:{PORT} --join {join} --name wsl-agent --identity-dir /root/mon/id")
-        cfg = {"fleet": {"server": f"http://127.0.0.1:{PORT}", "identity_dir": "/root/mon/id"}, "listen": {"bind": "0.0.0.0"}}
-        wsl("cat > /root/mon/agent.json <<'EOF'\n" + json.dumps(cfg) + "\nEOF")
-    assert start_agent() is not None
-    assert wait_until(lambda: api("GET", "/api/admin/agents")["items"][0]["online"], 20) is not None, "agent did not come online"
-    agent = api("GET", "/api/admin/agents")["items"][0]
-    AGENT_ID = agent["id"]
-    if fresh:
-        api("PATCH", "/api/admin/agents/" + AGENT_ID, {"public_host": WSLIP}, "application/merge-patch+json")
-        ident = api("POST", "/api/admin/probes", {"name": "home"})
-        json.dump(ident, open(os.path.join(WORK, "probe", "probe.json"), "w"))
-    PROBE.start()
-    assert wait_until(agent_view_probe_connected, 30) is not None, "probe did not connect"
-
-
-def agent_view():
-    return api("GET", "/api/admin/agents/" + AGENT_ID)
-
-
-def agent_view_probe_connected():
-    return agent_view()["probe_connected"]
-
-
-def agent_online():
-    return agent_view()["online"]
-
-
-def metrics(rng="2m", names="loss.total,loss.up,loss.down,rtt.p50,tcp.rtt.p50,tcp.rtt.max"):
-    s = api("GET", f"/api/public/lines/{AGENT_ID}/series?range={rng}&step=1s&metrics={names}")
-    return s["summary"]["metrics"], s["summary"]["samples"]
-
-
-def gaps(minutes=3):
-    """Buckets (10 s) in the last N minutes, after the first data, whose udp sample count is below half of the expected 100."""
-    s = api("GET", f"/api/public/lines/{AGENT_ID}/series?range={minutes}m&step=10s&metrics=n.udp_echo,n.tcp_echo")
-    udp = s["values"]["n.udp_echo"]
-    first = next((i for i, v in enumerate(udp) if v is not None and v >= 50), len(udp))  # buckets before the first data are not gaps
-    return [i for i, v in enumerate(udp[:-1]) if i > first and (v is None or v < 50)], len(udp)
-
-
-def stored_chunks():
-    return len(api("GET", f"/api/admin/raw/chunks?agent_id={AGENT_ID}&limit=1000")["items"])
-
-
-def healthy():
-    try:
-        return agent_online() and agent_view_probe_connected()
-    except Exception:
-        return False
-
-
-def ensure_healthy(timeout=60):
-    if not backend_up():
-        start_backend()
-    if not pids("moshiro-agent"):
-        start_agent()
-    if not PROBE.alive():
-        PROBE.start()
-    return wait_until(healthy, timeout)
-
-
-# ------------------------------------------------------------------ scenarios
-
-def scenario_init():
-    print("== init: cold starts in every order, bad configuration")
-    # 1. everything down, then start in the opposite order: probe -> agent -> backend (each must retry until its peer is up)
-    PROBE.stop("kill")
-    wsl("pkill -9 -x moshiro-agent; pkill -9 -x moshiro-backend; true")
-    time.sleep(1)
-    PROBE.start()
-    time.sleep(3)
-    start_agent()
-    time.sleep(3)
-    t0 = now()
-    start_backend()
-    took = wait_until(healthy, 90)
-    check("probe -> agent -> backend start order converges", took is not None, f"{took:.1f}s after the backend came up" if took else "never became healthy")
-    chunks_flow = wait_until(lambda: stored_chunks() > 0, 40)
-    check("data flows after the cold start", chunks_flow is not None)
-
-    # 2. backend first, then agent, then probe (the normal order) after a full stop
-    PROBE.stop("kill")
-    wsl("pkill -9 -x moshiro-agent; pkill -9 -x moshiro-backend; true")
-    start_backend()
-    start_agent()
-    PROBE.start()
-    took = wait_until(healthy, 60)
-    check("backend -> agent -> probe start order converges", took is not None, f"{took:.1f}s" if took else "")
-
-    # 3. the probe starts when its agent list is only a stale cache (agents.json wrong): bootstrap is still tried
-    # 4. bad configuration: clear errors and a non-zero exit code, no crash dump
-    bad = os.path.join(WORK, "bad.json")
-    open(bad, "w").write("{ not json")
-    r = subprocess.run([PROBE_EXE, "--config", bad], capture_output=True, text=True, timeout=20, creationflags=HIDDEN)
-    check("probe: malformed config exits 1 with a readable error", r.returncode == 1 and "startup failed" in r.stderr + r.stdout, (r.stderr + r.stdout).strip()[:120])
-    r = subprocess.run([PROBE_EXE, "--config", os.path.join(WORK, "missing.json")], capture_output=True, text=True, timeout=20, creationflags=HIDDEN)
-    check("probe: missing config exits 1 with the file name", r.returncode == 1 and "missing.json" in r.stderr + r.stdout, (r.stderr + r.stdout).strip()[:120])
-    out = wsl("cd /root/mon && echo '{ not json' > bad.json && ./moshiro-agent --config bad.json 2>&1; echo exit=$?", check=False)
-    check("agent: malformed config exits 1", "exit=1" in out and "startup failed" in out, out.strip()[-120:])
-    out = wsl("cd /root/mon && printf '{\"fleet\":{\"server\":\"http://127.0.0.1:1\",\"identity_dir\":\"/nonexistent\"}}' > noid.json && ./moshiro-agent --config noid.json 2>&1; echo exit=$?", check=False)
-    check("agent: no identity says to run fleet-enroll first", "exit=1" in out and "fleet-enroll" in out, out.strip()[-140:])
-    out = wsl(f"cd /root/mon && ./fleet-enroll --server http://127.0.0.1:{PORT} --join bogus --name x --identity-dir /tmp/enroll-x 2>&1; echo exit=$?", check=False)
-    check("fleet-enroll: a bad join token fails clearly (non-zero exit)", "exit=0" not in out, out.strip()[-140:])
-    out = wsl(f"cd /root/mon && ./fleet-enroll --server http://127.0.0.1:{PORT} --join bogus --name x --identity-dir /root/mon/id 2>&1; echo exit=$?", check=False)
-    check("fleet-enroll: refuses to overwrite an existing identity", "exit=0" not in out and "already enrolled" in out, out.strip()[-140:])
-    ensure_healthy()
-
-
-def scenario_graceful():
-    print("== graceful: SIGTERM / Ctrl+Break of each component, then restart")
-    ensure_healthy()
-    # probe (Ctrl+Break in its own console)
-    secs, code = PROBE.stop("graceful")
-    check("probe exits on Ctrl+Break within 5 s with code 0", secs is not None and secs < 5 and code == 0, f"{secs and round(secs, 2)}s code={code}")
-    seen = wait_until(lambda: not agent_view_probe_connected(), 10)
-    check("backend sees the probe disconnect", seen is not None, f"{seen and round(seen, 1)}s")
-    PROBE.start()
-    check("probe restarts and reconnects", wait_until(healthy, 30) is not None)
-
-    # agent
-    secs, code = stop_proc("moshiro-agent", "TERM")
-    log = wsl("tail -5 /root/mon/agent.log")
-    check("agent exits on SIGTERM within 5 s with code 0", secs is not None and secs < 5 and code == 0, f"{secs and round(secs, 2)}s code={code}")
-    check("agent logs a clean stop (no crash trace)", "stopping" in log and "Unhandled" not in log, log.splitlines()[-1][:100] if log else "")
-    seen = wait_until(lambda: not agent_online(), 10)
-    check("backend sees the agent go offline", seen is not None, f"{seen and round(seen, 1)}s")
-    start_agent()
-    check("agent restarts, reconnects to backend and probe", wait_until(healthy, 40) is not None)
-
-    # backend
-    secs, code = stop_proc("moshiro-backend", "TERM", 30)
-    log = wsl("tail -5 /root/mon/backend.log")
-    check("backend exits on SIGTERM within 15 s with code 0", secs is not None and secs < 15 and code == 0, f"{secs and round(secs, 2)}s code={code}")
-    check("backend log has no unhandled exception", "Unhandled" not in log and "Exception" not in log, log.replace(chr(10), ' | ')[-160:])
-    time.sleep(5)
-    start_backend()
-    check("backend restarts, agent and probe converge again", wait_until(healthy, 60) is not None)
-    ensure_healthy()
-
-
-def scenario_crash():
-    print("== crash: kill -9 / taskkill /F of each component, data durability")
-    ensure_healthy()
-    time.sleep(25)
-    before = stored_chunks()
-
-    # backend crash: the agent keeps its outbox; after restart nothing acknowledged is lost
-    t = now()
-    wsl("pkill -9 -x moshiro-backend")
-    time.sleep(25)  # agent outbox accumulates (2-3 chunk rounds)
-    start_backend()
-    back = wait_until(healthy, 60)
-    check("backend kill -9: agent reconnects and everything converges", back is not None, f"{back and round(back, 1)}s after restart")
-    time.sleep(20)
-    after = stored_chunks()
-    g, n = gaps(3)
-    check("backend kill -9: chunks buffered by the agent are delivered after the restart (no gap)", not g, f"gaps in {n} buckets: {g}; chunks {before}->{after}")
-
-    # agent crash
-    wsl("pkill -9 -x moshiro-agent")
-    time.sleep(5)
-    start_agent()
-    check("agent kill -9: recovers", wait_until(healthy, 40) is not None)
-
-    # probe crash
-    PROBE.stop("kill")
-    time.sleep(5)
-    PROBE.start()
-    check("probe kill: recovers", wait_until(healthy, 40) is not None)
-    time.sleep(25)
-    m, samples = metrics("30s")
-    check("after all three crashes the line reports correct data (no phantom loss)", (m.get("loss.up") or 0) < 1 and (m.get("loss.down") or 0) < 1, f"loss.up={m.get('loss.up')} loss.down={m.get('loss.down')}")
-
-    # crash in the middle of an on-demand test
-    test = api("POST", f"/api/admin/lines/{AGENT_ID}/tests", {"plan": "hy2_ladder", "params": {"size": 1250}})
-    time.sleep(12)
-    PROBE.stop("kill")
-    state = wait_until(lambda: api("GET", "/api/public/tests/" + test["id"])["status"] in ("failed", "aborted"), 30)
-    final = api("GET", "/api/public/tests/" + test["id"])
-    check("probe crash mid-test: the test ends (failed/aborted), it does not hang", state is not None, f"status={final['status']} error={final.get('error_code')}")
-    PROBE.start()
-    wait_until(healthy, 40)
-    test2 = api("POST", f"/api/admin/lines/{AGENT_ID}/tests", {"plan": "hy2_ladder", "params": {"size": 1250}})
-    done = wait_until(lambda: api("GET", "/api/public/tests/" + test2["id"])["status"] in ("succeeded", "failed", "aborted"), 90, 2)
-    check("a new test runs to completion after the crash (the agent is not stuck busy)", done is not None and api("GET", "/api/public/tests/" + test2["id"])["status"] == "succeeded", api("GET", "/api/public/tests/" + test2["id"])["status"])
-    ensure_healthy()
-
-
-def netem(device, spec, direction="egress", ports=(7443, 7444)):
-    """Impair only the data/control ports of the agent (so the rest of WSL is unaffected)."""
-    clear_netem(device)
-    if direction == "egress":
-        wsl(f"tc qdisc add dev {device} root handle 1: prio && tc qdisc add dev {device} parent 1:3 handle 30: netem {spec}")
-        for p in ports:
-            wsl(f"tc filter add dev {device} protocol ip parent 1:0 prio 3 u32 match ip sport {p} 0xffff flowid 1:3")
-    else:
-        wsl(f"tc qdisc add dev {device} handle ffff: ingress")
-        for p in ports:
-            action = "action drop" if spec == "drop" else ""
-            wsl(f"tc filter add dev {device} parent ffff: protocol ip u32 match ip dport {p} 0xffff {action}")
-
-
-def clear_netem(device):
-    wsl(f"tc qdisc del dev {device} root 2>/dev/null; tc qdisc del dev {device} ingress 2>/dev/null; true")
-
-
-def scenario_probe_link():
-    print("== probe<->agent link: flapping, long blackhole (half-open), reconnect storms")
-    ensure_healthy()
-    time.sleep(20)
-    mem0 = (PROBE.working_set_mb(), rss_mb("moshiro-agent"))
-
-    # 1. flapping: 3 s blackholes every 10 s, 8 times
-    for _ in range(8):
-        netem("eth0", "drop", "ingress")
-        time.sleep(3)
-        clear_netem("eth0")
-        time.sleep(7)
-    check("flapping link: probe still connected afterwards", wait_until(healthy, 30) is not None)
-
-    # 2. a 70 s blackhole: both ends must notice or survive, and recover once the path returns
-    t = now()
-    netem("eth0", "drop", "ingress")
-    time.sleep(70)
-    clear_netem("eth0")
-    back = wait_until(healthy, 120)
-    check("70 s blackhole: converges back to healthy", back is not None, f"{back and round(back, 1)}s after the path returned")
-    time.sleep(35)
-    m, _ = metrics("30s")
-    check("after the blackhole the line is measured correctly again", (m.get("loss.total") or 0) < 2, f"loss.total={m.get('loss.total')}")
-
-    # 3. reconnect storm: kill the agent 12 times in a row
-    for _ in range(12):
-        wsl("pkill -9 -x moshiro-agent; true")
-        time.sleep(0.5)
-        start_agent()
-        time.sleep(2)
-    check("12 agent restarts in a row: converges", wait_until(healthy, 60) is not None)
-    time.sleep(30)
-    g, n = gaps(1)
-    check("... and the data is flowing again", not g, f"gaps: {g}")
-
-    # 4. restart storm of the probe
-    for _ in range(8):
-        PROBE.stop("kill")
-        time.sleep(0.3)
-        PROBE.start()
-        time.sleep(2)
-    check("8 probe restarts in a row: converges", wait_until(healthy, 60) is not None)
-    time.sleep(40)
-    mem1 = (PROBE.working_set_mb(), rss_mb("moshiro-agent"))
-    check("memory stays bounded after the storms", mem1[0] < mem0[0] + 15 and mem1[1] < mem0[1] + 15, f"probe {mem0[0]:.1f}->{mem1[0]:.1f} MB, agent {mem0[1]:.1f}->{mem1[1]:.1f} MB")
-    ensure_healthy()
-
-
-def scenario_backend_link():
-    print("== agent<->backend link: outage buffering, replay, duplicates")
-    ensure_healthy()
-    time.sleep(30)
-    # an outage on loopback (tc on lo: the agent<->backend link) shorter than the buffering capacity
-    netem("lo", "drop", "ingress", ports=(PORT,))
-    t = now()
-    time.sleep(45)
-    clear_netem("lo")
-    check("45 s agent<->backend outage: agent stays up and reconnects", wait_until(agent_online, 90) is not None, f"{now() - t:.0f}s")
-    time.sleep(35)
-    g, n = gaps(4)
-    check("data produced during the outage is replayed from the agent's outbox (no gap)", not g, f"gaps in {n} buckets: {g}")
-
-    # duplicates: chunk ids are idempotent. The count of stored chunks grows by the expected 2 per 10 s, not more
-    c0 = stored_chunks()
-    time.sleep(30)
-    c1 = stored_chunks()
-    check("steady state: ~2 chunks / 10 s (udp + tcp), no duplicates", 4 <= c1 - c0 <= 8, f"{c1 - c0} chunks in 30 s")
-
-    # the backend is stopped for a while (graceful): the agent keeps collecting, delivers after
-    stop_proc("moshiro-backend", "TERM", 30)
-    time.sleep(40)
-    start_backend()
-    check("backend down 40 s then up: everything converges", wait_until(healthy, 90) is not None)
-    time.sleep(35)
-    g, n = gaps(4)
-    check("... with no gap in the data", not g, f"gaps: {g}")
-    ensure_healthy()
-
-
-def scenario_silent():
-    print("== silent death: a frozen process keeps its sockets open and says nothing (hung machine, lost power)")
-    ensure_healthy()
-    time.sleep(10)
-
-    # a frozen probe: the agent must notice and tell the backend, instead of keeping a dead session forever
-    PROBE.suspend()
-    t = now()
-    seen = wait_until(lambda: not agent_view_probe_connected(), 150, 1)
-    check("frozen probe: the backend learns the probe is gone", seen is not None, f"{seen and round(seen)}s" if seen else "still shown as connected after 150 s")
-    PROBE.resume()
-    check("... and when the probe thaws it comes back", wait_until(healthy, 90) is not None)
-
-    # a frozen agent: the probe must drop the session and keep retrying; the backend must mark the agent offline
-    wsl("pkill -STOP -x moshiro-agent")
-    seen_off = wait_until(lambda: not agent_online(), 150, 1)
-    check("frozen agent: the backend marks it offline", seen_off is not None, f"{seen_off and round(seen_off)}s" if seen_off else "still online after 150 s")
-    wsl("pkill -CONT -x moshiro-agent")
-    check("... and when the agent thaws everything converges", wait_until(healthy, 90) is not None)
-
-    # a frozen backend: the agent must notice (WebSocket ping timeout) and reconnect once the backend thaws
-    wsl("pkill -STOP -x moshiro-backend")
-    time.sleep(70)
-    wsl("pkill -CONT -x moshiro-backend")
-    check("frozen backend (70 s): converges after it thaws", wait_until(healthy, 90) is not None)
-    time.sleep(30)
-    g, n = gaps(3)
-    check("... and the data produced while it was frozen is delivered", not g, f"gaps: {g}")
-    ensure_healthy()
-
-
-def scenario_soak():
-    print("== soak: 8 minutes with random faults; memory, convergence and data continuity")
-    ensure_healthy()
-    time.sleep(20)
-    mem0 = (PROBE.working_set_mb(), rss_mb("moshiro-agent"), rss_mb("moshiro-backend"))
-    import random
-    rnd = random.Random(7)
-    end = now() + 8 * 60
-    faults = 0
-    while now() < end:
-        fault = rnd.choice(["blackhole", "loss", "agent_kill", "probe_kill", "none", "none"])
-        faults += fault != "none"
-        if fault == "blackhole":
-            netem("eth0", "drop", "ingress")
-            time.sleep(rnd.uniform(1, 6))
-            clear_netem("eth0")
-        elif fault == "loss":
-            netem("eth0", "delay 20ms 5ms loss 5%", "egress")
-            time.sleep(rnd.uniform(5, 12))
-            clear_netem("eth0")
-        elif fault == "agent_kill":
-            wsl("pkill -9 -x moshiro-agent; true")
-            time.sleep(1)
-            start_agent()
-        elif fault == "probe_kill":
-            PROBE.stop("kill")
-            time.sleep(0.5)
-            PROBE.start()
-        time.sleep(rnd.uniform(8, 20))
-    clear_netem("eth0")
-    check(f"after {faults} random faults everything converges", ensure_healthy(90) is not None)
-    time.sleep(45)
-    mem1 = (PROBE.working_set_mb(), rss_mb("moshiro-agent"), rss_mb("moshiro-backend"))
-    check("memory bounded", mem1[0] < mem0[0] + 15 and mem1[1] < mem0[1] + 20 and mem1[2] < mem0[2] + 40, f"probe {mem0[0]:.0f}->{mem1[0]:.0f}, agent {mem0[1]:.0f}->{mem1[1]:.0f}, backend {mem0[2]:.0f}->{mem1[2]:.0f} MB")
-    m, _ = metrics("30s")
-    check("the final minute is measured correctly", (m.get("loss.up") or 0) < 1 and (m.get("loss.down") or 0) < 1, f"{m.get('loss.up')} {m.get('loss.down')}")
-
-
-SCENARIOS = {
-    "init": scenario_init,
-    "graceful": scenario_graceful,
-    "crash": scenario_crash,
-    "probe_link": scenario_probe_link,
-    "backend_link": scenario_backend_link,
-    "silent": scenario_silent,
-    "soak": scenario_soak,
-}
-
-
-def main():
-    names = sys.argv[1:] or list(SCENARIOS)
-    print(f"wsl ip {WSLIP}")
-    setup(fresh=True)
-    for name in names:
-        try:
-            SCENARIOS[name]()
-        except Exception as ex:
-            check(f"{name} ran to completion", False, repr(ex)[:300])
-    PROBE.stop("kill")
-    wsl("pkill -9 -x moshiro-agent; pkill -9 -x moshiro-backend; tc qdisc del dev eth0 root 2>/dev/null; tc qdisc del dev eth0 ingress 2>/dev/null; tc qdisc del dev lo ingress 2>/dev/null; true")
-    failed = [r for r in RESULTS if not r[1]]
-    print(f"\n{len(RESULTS) - len(failed)}/{len(RESULTS)} checks passed")
-    for name, _, detail in failed:
-        print(f"FAILED: {name} - {detail}")
-    sys.exit(1 if failed else 0)
-
-
-if __name__ == "__main__":
-    main()

scripts/test-openwrt-updater.py

@@ -44,11 +44,11 @@ elif name == 'jsonfilter':
     args = sys.argv[1:]
     data = json.loads(pathlib.Path(args[args.index('-i') + 1]).read_text() if '-i' in args else sys.stdin.read())
     query = args[args.index('-e') + 1]
-    value = data['moshiro-probe']['instances']['default']['pid'] if query.startswith('@[') else data.get(query[2:])
+    value = data['moshiro-agent']['instances']['default']['pid'] if query.startswith('@[') else data.get(query[2:])
     if value is None: sys.exit(1)
     print(value)
 elif name == 'ubus':
-    print(json.dumps({'moshiro-probe': {'instances': {'default': {'pid': 1234}}}}))
+    print(json.dumps({'moshiro-agent': {'instances': {'default': {'pid': 1234}}}}))
 elif name == 'service':
     with (base / 'restarts').open('a') as f: f.write('restart\n')
     if os.environ.get('HEALTH') != 'good': sys.exit(0)
@@ -93,7 +93,7 @@ class UpdaterTests(unittest.TestCase):
 
     def publish(self, *, seq=100, version="1.0.1-dev.1", members=None, changes=None):
         payload = io.BytesIO()
-        members = members or {name: b"executable\n" for name in ("moshiro-probe", "fleet-enroll", "openwrt.sh", "probe.init", "updater.init")}
+        members = members or {name: b"executable\n" for name in ("moshiro-agent", "fleet-enroll", "openwrt.sh", "agent.init", "updater.init")}
         with tarfile.open(fileobj=payload, mode="w:gz") as archive:
             for name, content in members.items():
                 entry = tarfile.TarInfo(name)
@@ -101,11 +101,11 @@ class UpdaterTests(unittest.TestCase):
                 archive.addfile(entry, io.BytesIO(content))
         data = payload.getvalue()
         self.hash = hashlib.sha256(data).hexdigest()
-        package = f"probe-linux-musl-arm64-{self.hash}.tar.gz"
+        package = f"agent-linux-musl-arm64-{self.hash}.tar.gz"
         version_dir = self.base / "feed" / version
         version_dir.mkdir(exist_ok=True)
         (version_dir / package).write_bytes(data)
-        self.manifest = dict(schema=1, component="probe", channel="dev", rid="linux-musl-arm64", version=version,
+        self.manifest = dict(schema=1, component="agent", channel="dev", rid="linux-musl-arm64", version=version,
                              package=package, sha256=self.hash, size=len(data), seq=seq)
         self.manifest.update(changes or {})
         self.sign()
@@ -144,7 +144,7 @@ class UpdaterTests(unittest.TestCase):
     def test_ci_publisher_produces_a_signed_package_the_updater_can_install(self):
         binaries = self.base / "binaries"
         binaries.mkdir()
-        for name in ("moshiro-probe", "fleet-enroll"):
+        for name in ("moshiro-agent", "fleet-enroll"):
             (binaries / name).write_bytes(b"native executable fixture\n")
         env = dict(self.env, CI_COMMIT_TAG="", CI_PIPELINE_EVENT="push", CI_COMMIT_BRANCH="main", CI_COMMIT_MESSAGE="[CI] fixture", GITEA_BASE="https://test.invalid",
                    GITEA_TOKEN="test-token", RELEASE_SIGNING_KEY=self.key.read_text())