merge feat/v221: windows probe degradation + agent heartbeat probe + update-check fallback + procd/OpenRC script fixes + probe traffic GB unification + public remark name (V2.2.1)

This commit is contained in:
Muse
2026-10-23 15:24:55 +00:00
19 changed files with 794 additions and 42 deletions
+3 -3
View File
@@ -307,13 +307,13 @@ export default function NodeList({
</Tooltip>
) : (
<Space>
<Tooltip title={t('pages.nodes.probe')}>
<Tooltip title={t('pages.nodes.testConnection')}>
<Button
type="text"
size="small"
style={{ fontSize: 16 }}
icon={<ThunderboltOutlined />}
aria-label={t('pages.nodes.probe')}
aria-label={t('pages.nodes.testConnection')}
onClick={() => onProbe(record)}
/>
</Tooltip>
@@ -716,7 +716,7 @@ export default function NodeList({
key: 'probe',
label: (
<>
<ThunderboltOutlined /> {t('pages.nodes.probe')}
<ThunderboltOutlined /> {t('pages.nodes.testConnection')}
</>
),
onClick: () => onProbe(record),
+30 -8
View File
@@ -24,12 +24,20 @@ interface ProbeMetaWire {
ipv4Pin: string;
ipv6Pin: string;
// V2.0.1: ledger corrections in GB (either sign) and the counted
// interface pin ('' = automatic selection on the agent).
// interface pin ('' = automatic selection on the agent). Since
// V2.2.1 the corrections are one-shot deltas: the inputs below are
// local state (never prefilled from the wire), a save submits them
// once, and they clear back to empty on success.
trafficCalUpGb: number;
trafficCalDownGb: number;
statIface: string;
}
// GB entry converts on the binary convention every byte formatter in
// the panel and on the public page uses (V2.2.1: was decimal 1e9,
// which made a 3000 GB limit render as 2.73 TB instead of 2.93 TB).
const BYTES_PER_GB = 1024 ** 3;
const JSON_HEADERS = { headers: { 'Content-Type': 'application/json' } };
const NAMED_CYCLES = [
@@ -59,6 +67,9 @@ export default function ProbeMetaSection({ nodeId }: { nodeId: number }) {
// in NodeFormModal); visibility derives during render.
const [loaded, setLoaded] = useState<{ nodeId: number; meta: ProbeMetaWire } | null>(null);
const [trafficGb, setTrafficGb] = useState(0);
// One-shot calibration deltas (GB, either sign): null = empty box.
const [calUpGb, setCalUpGb] = useState<number | null>(null);
const [calDownGb, setCalDownGb] = useState<number | null>(null);
const [saving, setSaving] = useState(false);
useEffect(() => {
@@ -68,7 +79,7 @@ export default function ProbeMetaSection({ nodeId }: { nodeId: number }) {
}).then((msg) => {
if (!cancelled && msg?.success && msg.obj) {
setLoaded({ nodeId, meta: msg.obj });
setTrafficGb(msg.obj.trafficLimit / 1e9);
setTrafficGb(msg.obj.trafficLimit / BYTES_PER_GB);
}
});
return () => {
@@ -116,14 +127,23 @@ export default function ProbeMetaSection({ nodeId }: { nodeId: number }) {
try {
const payload: ProbeMetaWire = {
...meta,
trafficLimit: Math.round(trafficGb * 1e9),
trafficLimit: Math.round(trafficGb * BYTES_PER_GB),
trafficCalUpGb: calUpGb ?? 0,
trafficCalDownGb: calDownGb ?? 0,
};
const msg = await HttpUtil.put(`/panel/api/agent/probe-meta/${nodeId}`, payload, {
silent: true,
...JSON_HEADERS,
});
if (msg?.success) {
setLoaded({ nodeId, meta: payload });
// One-shot: the submitted corrections are folded in server-
// side; the boxes clear so a later save resubmits nothing.
setLoaded({
nodeId,
meta: { ...payload, trafficCalUpGb: 0, trafficCalDownGb: 0 },
});
setCalUpGb(null);
setCalDownGb(null);
messageApi.success(t('pages.nodes.probe.saved'));
} else {
messageApi.error(msg?.msg || t('pages.nodes.probe.saveFailed'));
@@ -249,9 +269,10 @@ export default function ProbeMetaSection({ nodeId }: { nodeId: number }) {
<Col xs={24} md={8}>
<Field label={t('pages.nodes.probe.calUp')}>
<InputNumber
value={meta.trafficCalUpGb}
value={calUpGb}
step={0.1}
onChange={(v) => set('trafficCalUpGb', v ?? 0)}
placeholder="0"
onChange={(v) => setCalUpGb(v)}
style={{ width: '100%' }}
/>
</Field>
@@ -259,9 +280,10 @@ export default function ProbeMetaSection({ nodeId }: { nodeId: number }) {
<Col xs={24} md={8}>
<Field label={t('pages.nodes.probe.calDown')}>
<InputNumber
value={meta.trafficCalDownGb}
value={calDownGb}
step={0.1}
onChange={(v) => set('trafficCalDownGb', v ?? 0)}
placeholder="0"
onChange={(v) => setCalDownGb(v)}
style={{ width: '100%' }}
/>
</Field>
+61 -4
View File
@@ -11,7 +11,7 @@
#
# 测试用环境变量(一般无需设置):
# XUI_ROOT 安装路径前缀(假根目录验证用),默认空 = 真实路径
# XUI_INIT_SYSTEM 强制 init 系统: systemd | openrc | none,默认 auto 探测
# XUI_INIT_SYSTEM 强制 init 系统: systemd | openrc | procd | none,默认 auto 探测
# XUI_URL agent 二进制下载地址覆盖
#
# 二进制命名(P12 改名):agent 独立装为 /usr/local/x-ui/xui-agent,不再与主控共用
@@ -220,7 +220,7 @@ stop_legacy_agent() {
detect_init_system() {
case "${XUI_INIT_SYSTEM:-auto}" in
systemd|openrc|none) echo "${XUI_INIT_SYSTEM}"; return;;
systemd|openrc|procd|none) echo "${XUI_INIT_SYSTEM}"; return;;
esac
if [ -d /run/systemd/system ] && command -v systemctl >/dev/null 2>&1; then
echo systemd; return
@@ -228,6 +228,9 @@ detect_init_system() {
if command -v rc-update >/dev/null 2>&1 && command -v rc-service >/dev/null 2>&1; then
echo openrc; return
fi
if [ -f /etc/rc.common ] && command -v procd >/dev/null 2>&1; then
echo procd; return
fi
echo none
}
@@ -286,10 +289,14 @@ command_args="agent"
# supervise-daemon 守护(第 23 条):进程崩溃后按 respawn 参数自动拉起,
# 对齐 systemd 支的 Restart=on-failure + RestartSec=5;旧版 command_background
# 只负责后台化、崩溃不重启。respawn_max=0 = 不限次(同 systemd 的持续重试)。
# respawn_period 必须为正数:supervise-daemon 启动时校验该值,0 直接报
# "Invalid respawn-period value '0'" 拒绝拉起(V2.2.1 ⑤,Alpine 真机实证);
# 取 1800(真机上其他 supervise-daemon 服务同值)。max=0 不限次时 period
# 仅是计数窗口,不影响持续重试语义。
supervisor="supervise-daemon"
respawn_delay=5
respawn_max=0
respawn_period=0
respawn_period=1800
pidfile="/run/x-ui-agent.pid"
output_log="/var/log/x-ui-agent.log"
error_log="/var/log/x-ui-agent.log"
@@ -322,9 +329,58 @@ EOF
fi
}
# 以 procd 服务方式安装并启动(开机自启,OpenWrt)
start_via_procd() {
local init_dir="${ROOT}/etc/init.d"
mkdir -p "$init_dir"
# 服务脚本与 2026-10-23 AX6600 真机手工验证版逐字一致(V2.2.1 ④
# 收编):路径按脚本变量展开,XUI_ROOT 为空时即验证版的
# /usr/local/x-ui/xui-agent 与 /etc/x-ui/agent.json。procd 托管
# + respawn 守护,stdout/stderr 进 syslog(logread 查看)。
cat > "${init_dir}/xui-agent" <<EOF
#!/bin/sh /etc/rc.common
START=99
STOP=10
USE_PROCD=1
start_service() {
procd_open_instance
procd_set_param command ${AGENT_BIN} agent -config ${CONF_DIR}/agent.json
procd_set_param respawn 3600 5 5
procd_set_param stdout 1
procd_set_param stderr 1
procd_close_instance
}
EOF
chmod +x "${init_dir}/xui-agent"
info "已生成 procd 服务文件: ${init_dir}/xui-agent"
if [ -n "$ROOT" ]; then
info "(XUI_ROOT 模式:跳过 /etc/init.d 实际操作)"
return 0
fi
/etc/init.d/xui-agent stop 2>/dev/null || true
stop_legacy_agent
pkill -f "${AGENT_BIN} agent" 2>/dev/null || true; sleep 1
/etc/init.d/xui-agent enable
/etc/init.d/xui-agent start
sleep 2
if pgrep -f "${AGENT_BIN} agent" >/dev/null 2>&1 || ps w 2>/dev/null | grep "[x]ui-agent agent" >/dev/null 2>&1; then
info "agent 已由 procd 托管启动(开机自启)"
echo ""
echo " 主控: $MASTER"
echo " 节点名: $NAME"
echo " 日志: logread -f"
echo " agent 会在 30 秒内从主控拉取配置"
else
error "agent 服务启动失败,查看: logread | tail -n 50"
fi
}
# 无服务管理器时回落 nohup(无开机自启,明确警告)
start_via_nohup() {
warn "未检测到 systemd / OpenRC,退回 nohup 方式启动"
# BusyBox 精简系统可能连 nohup 都没有(AX6600 实证):先讲清再退,
# 不要假装启动、走到最后才报一个看不懂的失败。
command -v nohup >/dev/null 2>&1 || error "未检测到 systemd / OpenRC / procd,且本机无 nohup 命令,无法启动 agent"
warn "未检测到 systemd / OpenRC / procd,退回 nohup 方式启动"
warn "注意:nohup 方式没有开机自启,机器重启后 agent 不会自动恢复,需手动重跑本脚本!"
stop_legacy_agent
pkill -f "${AGENT_BIN} agent" 2>/dev/null || true; sleep 1
@@ -348,5 +404,6 @@ info "启动 agent..."
case "$(detect_init_system)" in
systemd) start_via_systemd;;
openrc) start_via_openrc;;
procd) start_via_procd;;
*) start_via_nohup;;
esac
+56
View File
@@ -0,0 +1,56 @@
package agent
import (
"sync"
"github.com/mhsanaei/3x-ui/v3/internal/logger"
)
// This file holds the probe collector's degrade-don't-die plumbing:
// the per-source independence discipline as reusable mechanics,
// shared by every platform sampler but exercised today by the Windows
// sampler (the Linux sampler reads /proc files, whose failure mode is
// an error return, not a panic).
//
// Two layers:
//
// - source gates: an optional OS facility (a Win32 procedure, say)
// is probed once; when it is missing the source is skipped —
// zeroed — with a single warning, never re-probed, and never
// reached through a call path that panics on a missing export the
// way LazyProc.Call does when it must-finds a procedure;
// - guarded calls: a panicking source is recovered, zeroed and
// logged, so one bad source can never take the agent process (and
// every other source with it) down.
// sourceGate caches one availability probe. The zero value is ready.
type sourceGate struct {
once sync.Once
available bool
}
// enabled runs probe on first use and reports whether the source may
// be used. probe returns nil when the facility is present. A failed
// probe disables the source for the process lifetime and logs once.
func (g *sourceGate) enabled(name string, probe func() error) bool {
g.once.Do(func() {
if err := probe(); err != nil {
logger.Warningf("probe: %s unavailable, source disabled: %v", name, err)
return
}
g.available = true
})
return g.available
}
// guardedCall runs fn and converts a panic into the zero value plus a
// warning, so one failing metrics source degrades alone.
func guardedCall[T any](name string, fn func() T) (out T) {
defer func() {
if r := recover(); r != nil {
logger.Warningf("probe: %s panicked, source zeroed: %v", name, r)
out = *new(T)
}
}()
return fn()
}
+63
View File
@@ -0,0 +1,63 @@
package agent
import (
"errors"
"testing"
)
// The degrade-don't-die mechanics behind V2.2.1 ①: on the trimmed
// Tianyi cloud-PC image the Windows sampler died on its first
// collection because a missing iphlpapi export made LazyProc.Call
// panic. These tests pin the two layers that replace that failure
// mode — a failed availability probe disables the source (probed
// once, not per poll), and a panicking source is zeroed, never
// propagated.
func TestSourceGateDisablesFailedProbeOnce(t *testing.T) {
var g sourceGate
calls := 0
probe := func() error {
calls++
return errors.New("procedure not found")
}
for i := range 3 {
if g.enabled("test-source", probe) {
t.Fatalf("call %d: gate enabled despite failed probe", i)
}
}
if calls != 1 {
t.Fatalf("probe ran %d times, want exactly 1 (cached verdict)", calls)
}
}
func TestSourceGateEnablesSuccessfulProbeOnce(t *testing.T) {
var g sourceGate
calls := 0
probe := func() error {
calls++
return nil
}
for i := range 3 {
if !g.enabled("test-source", probe) {
t.Fatalf("call %d: gate disabled despite successful probe", i)
}
}
if calls != 1 {
t.Fatalf("probe ran %d times, want exactly 1 (cached verdict)", calls)
}
}
func TestGuardedCallRecoversPanicToZero(t *testing.T) {
got := guardedCall("test-source", func() uint64 {
panic("Failed to find FreeMibIfTable2 procedure in iphlpapi.dll")
})
if got != 0 {
t.Fatalf("guardedCall after panic = %d, want zero value", got)
}
}
func TestGuardedCallPassesValueThrough(t *testing.T) {
if got := guardedCall("test-source", func() string { return "ok" }); got != "ok" {
t.Fatalf("guardedCall = %q, want %q", got, "ok")
}
}
+82 -15
View File
@@ -35,6 +35,14 @@ package agent
// degradation discipline Load stays zero and TCP/UDP stay 0
// until a wrapper exists;
// - every call fails soft to zeros, like an unreadable /proc file.
// Optional procedures are resolved through source gates
// (probe_guard.go): LazyProc.Call must-finds and PANICS on a
// missing export, and trimmed images do ship DLLs without the
// full surface — the Tianyi cloud PC's iphlpapi.dll has no
// FreeMibIfTable2, which killed the V2.2.0 agent on its first
// collection (V2.2.1 ①). A missing procedure now disables just
// its source, and collectMetrics additionally guards each source
// so a panic anywhere degrades that source alone.
import (
"net"
@@ -50,15 +58,35 @@ import (
"golang.org/x/sys/windows/registry"
)
// kernel32 procs x/sys/windows v0.47.0 does not wrap.
// kernel32 procs x/sys/windows v0.47.0 does not wrap, plus the
// iphlpapi pair this file calls directly (gated below, so x/sys'
// own GetIfTable2Ex wrapper — whose Addr() panics the same way on a
// missing export — is deliberately not used).
var (
kernel32DLL = windows.NewLazySystemDLL("kernel32.dll")
procGetSystemTimes = kernel32DLL.NewProc("GetSystemTimes")
procGlobalMemoryStatus = kernel32DLL.NewProc("GlobalMemoryStatusEx")
iphlpapiDLL = windows.NewLazySystemDLL("iphlpapi.dll")
procGetIfTable2Ex = iphlpapiDLL.NewProc("GetIfTable2Ex")
procFreeMibIfTable2 = iphlpapiDLL.NewProc("FreeMibIfTable2")
)
// Availability gates for the optional procedures above. A LazyProc
// must never be Call()ed before its gate has proven the export
// exists: Call must-finds, so a missing procedure is a process-killing
// panic, not an error.
var (
gateSystemTimes sourceGate
gateMemoryStatus sourceGate
gateIfTable sourceGate
)
// procProbe adapts a LazyProc to a source-gate probe: Find resolves
// the export WITHOUT the mustFind panic baked into Call/Addr.
func procProbe(p *windows.LazyProc) func() error {
return func() error { return p.Find() }
}
func sampleFacts() *ProbeFacts {
memTotal, _, swapTotal, _ := winMemory()
diskTotal, _ := winDiskUsage()
@@ -80,12 +108,26 @@ func sampleFacts() *ProbeFacts {
}
}
// memSnapshot / diskSnapshot pack the multi-value samplers so each
// can run as one guarded source.
type memSnapshot struct{ total, used, swapTotal, swapUsed uint64 }
type diskSnapshot struct{ total, used uint64 }
func collectMetrics() *ProbeMetrics {
probeColl.mu.Lock()
defer probeColl.mu.Unlock()
memTotal, memUsed, swapTotal, swapUsed := winMemory()
diskTotal, diskUsed := winDiskUsage()
rows := winIfaceRows()
// Every source runs guarded: a panic in one (a Win32 surface the
// gates did not foresee) zeroes that source alone instead of
// killing the agent on its first collection.
mem := guardedCall("memory", func() memSnapshot {
total, used, swapTotal, swapUsed := winMemory()
return memSnapshot{total, used, swapTotal, swapUsed}
})
disk := guardedCall("disk", func() diskSnapshot {
total, used := winDiskUsage()
return diskSnapshot{total, used}
})
rows := guardedCall("interfaces", winIfaceRows)
all := make([]netCounters, 0, len(rows))
counted := make([]netCounters, 0, len(rows))
for _, r := range rows {
@@ -107,23 +149,25 @@ func collectMetrics() *ProbeMetrics {
}
rxRate, txRate := probeColl.netRate(counted, time.Now())
return &ProbeMetrics{
BootID: bootEpoch(winMachineGUID(), names),
Uptime: uint64(windows.DurationSinceBoot().Seconds()),
CPU: probeColl.cpuPercent(winCPUSample()),
BootID: bootEpoch(winMachineGUID(), names),
Uptime: guardedCall("uptime", func() uint64 {
return uint64(windows.DurationSinceBoot().Seconds())
}),
CPU: probeColl.cpuPercent(guardedCall("cpu", winCPUSample)),
Load: [3]float32{},
MemTotal: memTotal,
MemUsed: memUsed,
SwapTotal: swapTotal,
SwapUsed: swapUsed,
DiskTotal: diskTotal,
DiskUsed: diskUsed,
MemTotal: mem.total,
MemUsed: mem.used,
SwapTotal: mem.swapTotal,
SwapUsed: mem.swapUsed,
DiskTotal: disk.total,
DiskUsed: disk.used,
NetRxTotal: rxTotal,
NetTxTotal: txTotal,
NetRx: rxRate,
NetTx: txRate,
TCP: 0, // no exported x/sys wrapper (see header)
UDP: 0, // no exported x/sys wrapper (see header)
Procs: winProcCount(),
Procs: guardedCall("processes", winProcCount),
}
}
@@ -212,6 +256,9 @@ func filetime100ns(ft windows.Filetime) uint64 {
// winCPUSample reads the cumulative system times as one cpuSample:
// total = kernel + user (kernel already includes idle, like jiffies).
func winCPUSample() *cpuSample {
if !gateSystemTimes.enabled("GetSystemTimes", procProbe(procGetSystemTimes)) {
return nil
}
var idle, kernel, user windows.Filetime
r, _, _ := procGetSystemTimes.Call(
uintptr(unsafe.Pointer(&idle)),
@@ -248,6 +295,9 @@ func usedFromTotalAvail(total, avail uint64) uint64 {
}
func winMemory() (total, used, swapTotal, swapUsed uint64) {
if !gateMemoryStatus.enabled("GlobalMemoryStatusEx", procProbe(procGlobalMemoryStatus)) {
return 0, 0, 0, 0
}
var ms memoryStatusEx
ms.Length = uint32(unsafe.Sizeof(ms))
r, _, _ := procGlobalMemoryStatus.Call(uintptr(unsafe.Pointer(&ms)))
@@ -359,8 +409,25 @@ func winIfaceCounts(r winIfaceRow) bool {
// derives both the all-interfaces list (for the operator pin) and the
// counted set from it.
func winIfaceRows() []winIfaceRow {
// The get/free pair is gated as ONE source: calling GetIfTable2Ex
// without FreeMibIfTable2 would leak the system-allocated table
// on every poll, and either export missing must degrade the
// source to zero rows instead of panicking — the trimmed Tianyi
// cloud-PC image ships an iphlpapi.dll without FreeMibIfTable2,
// and the V2.2.0 sampler's unguarded LazyProc.Call died on it.
if !gateIfTable.enabled("GetIfTable2Ex/FreeMibIfTable2 (iphlpapi)", func() error {
if err := procGetIfTable2Ex.Find(); err != nil {
return err
}
return procFreeMibIfTable2.Find()
}) {
return nil
}
var table *windows.MibIfTable2
if err := windows.GetIfTable2Ex(windows.MibIfTableNormal, &table); err != nil || table == nil {
r, _, _ := procGetIfTable2Ex.Call(
uintptr(windows.MibIfTableNormal),
uintptr(unsafe.Pointer(&table)))
if r != 0 || table == nil {
return nil
}
defer procFreeMibIfTable2.Call(uintptr(unsafe.Pointer(table)))
+19 -4
View File
@@ -52,7 +52,10 @@ type probeMetaForm struct {
IPv4Pin string `json:"ipv4Pin"`
IPv6Pin string `json:"ipv6Pin"`
// V2.0.1: traffic calibration (GB, either sign) and the counted
// interface pin ("" = automatic).
// interface pin ("" = automatic). Since V2.2.1 ⑥b the calibration
// fields are ONE-SHOT deltas: a non-zero value is folded into the
// stored standing correction on save (see probe.MergeCalDelta);
// zero/absent applies nothing.
TrafficCalUpGB float64 `json:"trafficCalUpGb"`
TrafficCalDownGB float64 `json:"trafficCalDownGb"`
StatIface string `json:"statIface"`
@@ -75,8 +78,11 @@ func metaToForm(m probe.Meta) probeMetaForm {
CountryPin: m.CountryPin,
IPv4Pin: m.IPv4Pin,
IPv6Pin: m.IPv6Pin,
TrafficCalUpGB: m.TrafficCalUpGB,
TrafficCalDownGB: m.TrafficCalDownGB,
// One-shot delta inputs (V2.2.1 ⑥b): the form's calibration
// boxes always start empty; the standing correction folded
// into the stored row is never echoed back into them.
TrafficCalUpGB: 0,
TrafficCalDownGB: 0,
StatIface: m.StatIface,
}
}
@@ -150,7 +156,16 @@ func (pc *ProbeMetaController) putMeta(c *gin.Context) {
jsonMsg(c, msg, errRequest)
return
}
if err := probe.SaveMeta(form.toMeta(int64(agent.Id))); err != nil {
// Calibration is one-shot (V2.2.1 ⑥b): fold the submitted deltas
// into the stored standing correction; all other fields replace.
meta := form.toMeta(int64(agent.Id))
if existing, found, err := probe.MetaFor(int64(agent.Id)); err != nil {
jsonMsg(c, "save probe meta failed", err)
return
} else if found {
meta = probe.MergeCalDelta(existing, meta)
}
if err := probe.SaveMeta(meta); err != nil {
jsonMsg(c, "save probe meta failed", err)
return
}
+3 -1
View File
@@ -323,7 +323,9 @@ func TestProbePublicContract(t *testing.T) {
assertKeys(t, keySet(t, nodeBody), anonNodeKeys)
var node map[string]any
_ = json.Unmarshal(nodeBody, &node)
if node["id"] != float64(pubID) || node["name"] != "PubNode" || node["online"] != true ||
// V2.2.1 ⑦: the public view's name is the public remark when one
// is set (the seed sets "shown publicly"), not the node name.
if node["id"] != float64(pubID) || node["name"] != "shown publicly" || node["online"] != true ||
node["month_rx"] != float64(33) || node["month_used"] != float64(77) ||
node["total_tx"] != float64(222) || node["price"] != float64(10) {
t.Fatalf("anon node values wrong: %v", node)
+67
View File
@@ -1234,7 +1234,74 @@ func (s *NodeService) AggregateNodeMetric(id int, metric string, bucketSeconds i
return nodeMetrics.aggregate(nodeMetricKey(id, metric), bucketSeconds, maxPoints)
}
// agentOnlineWindow is how fresh an agent heartbeat must be for its
// node to read online — the same 180s the heartbeat job expires agent
// nodes by (agentHeartbeatStaleAfter in internal/web/job: max(3
// heartbeat intervals, 120s), so two lost heartbeats don't flap).
const agentOnlineWindow = 180 * time.Second
var errAgentOffline = errors.New("agent node offline: no fresh heartbeat")
// agentHeartbeatVerdict is the pure decision behind the agent-node
// probe: a fresh stamp reads online with the heartbeat AGE as the
// latency (there is no dial to time); a stale or never-seen stamp
// reads offline with LatencyMs -1. LastError stays empty either way —
// the answer comes from the heartbeat, not from a dial failure.
func agentHeartbeatVerdict(lastSeenMs int64, now time.Time) (HeartbeatPatch, bool) {
patch := HeartbeatPatch{LatencyMs: -1}
if lastSeenMs <= 0 {
return patch, false
}
patch.LastHeartbeat = lastSeenMs / 1000
if age := now.Sub(time.UnixMilli(lastSeenMs)); age <= agentOnlineWindow {
patch.LastHeartbeat = now.Unix()
patch.LatencyMs = int(age / time.Millisecond)
return patch, true
}
return patch, false
}
// agentNodeProbe answers a probe for an agent-managed node from the
// bound agent's last_seen stamp (Unix ms, the unit the agent
// controller stamps). ok=false means the node has no bound agent — or
// the lookup failed — and the caller takes the dial path. The stored
// telemetry fields carry through unchanged (same as the heartbeat
// job's agent-expiry patch); only the error is dropped, so a ⚡ click
// also heals the bogus dial errors recorded before this change.
func (s *NodeService) agentNodeProbe(n *model.Node) (patch HeartbeatPatch, err error, ok bool) {
var lastSeen *int64 // MAX() over zero rows is NULL
qerr := database.GetDB().Model(&model.AgentNode{}).
Where("node_id = ?", n.Id).
Select("MAX(last_seen)").
Scan(&lastSeen).Error
if qerr != nil || lastSeen == nil {
return HeartbeatPatch{}, nil, false
}
patch, online := agentHeartbeatVerdict(*lastSeen, time.Now())
patch.XrayVersion = n.XrayVersion
patch.PanelVersion = n.PanelVersion
patch.Guid = n.Guid
patch.CpuPct = n.CpuPct
patch.MemPct = n.MemPct
patch.UptimeSecs = n.UptimeSecs
patch.NetUp = n.NetUp
patch.NetDown = n.NetDown
patch.XrayState = n.XrayState
patch.XrayError = n.XrayError
if !online {
return patch, errAgentOffline, true
}
return patch, nil, true
}
func (s *NodeService) Probe(ctx context.Context, n *model.Node) (HeartbeatPatch, error) {
// Agent-managed nodes have no panel to dial — their address row
// is a placeholder ("agent") — so a probe answers from the agent
// heartbeat instead of dialing a dead URL and persisting a
// LastError the agent heartbeat never clears (V2.2.1 ②).
if patch, err, ok := s.agentNodeProbe(n); ok {
return patch, err
}
proxyURL := ""
if n.OutboundTag != "" {
if mgr := runtime.GetManager(); mgr != nil {
@@ -0,0 +1,70 @@
package service
import (
"testing"
"time"
)
// V2.2.1 ②: the ⚡ probe for an agent-managed node is synthesized
// from the agent's heartbeat stamp instead of dialing the placeholder
// address. These tests pin the pure verdict: fresh stamp → online
// with the heartbeat age as latency; stale / never-seen → offline
// with LatencyMs -1; LastError empty in every case (a dial error must
// never be recorded for an agent node again).
func TestAgentHeartbeatVerdictFreshIsOnlineWithAgeLatency(t *testing.T) {
now := time.Unix(1_700_000_000, 0)
lastSeenMs := now.Add(-30 * time.Second).UnixMilli()
patch, online := agentHeartbeatVerdict(lastSeenMs, now)
if !online {
t.Fatal("30s-old heartbeat must read online")
}
if patch.LatencyMs != 30_000 {
t.Fatalf("LatencyMs = %d, want heartbeat age 30000", patch.LatencyMs)
}
if patch.LastHeartbeat != now.Unix() {
t.Fatalf("LastHeartbeat = %d, want now %d", patch.LastHeartbeat, now.Unix())
}
if patch.LastError != "" {
t.Fatalf("LastError = %q, want empty", patch.LastError)
}
}
func TestAgentHeartbeatVerdictWindowEdge(t *testing.T) {
now := time.Unix(1_700_000_000, 0)
if _, online := agentHeartbeatVerdict(now.Add(-agentOnlineWindow).UnixMilli(), now); !online {
t.Fatal("heartbeat exactly at the window edge must read online")
}
if _, online := agentHeartbeatVerdict(now.Add(-agentOnlineWindow-time.Second).UnixMilli(), now); online {
t.Fatal("heartbeat past the window must read offline")
}
}
func TestAgentHeartbeatVerdictStaleIsOffline(t *testing.T) {
now := time.Unix(1_700_000_000, 0)
lastSeenMs := now.Add(-10 * time.Minute).UnixMilli()
patch, online := agentHeartbeatVerdict(lastSeenMs, now)
if online {
t.Fatal("10min-old heartbeat must read offline")
}
if patch.LatencyMs != -1 {
t.Fatalf("LatencyMs = %d, want -1", patch.LatencyMs)
}
if patch.LastHeartbeat != lastSeenMs/1000 {
t.Fatalf("LastHeartbeat = %d, want the truthful last-seen %d", patch.LastHeartbeat, lastSeenMs/1000)
}
if patch.LastError != "" {
t.Fatalf("LastError = %q, want empty", patch.LastError)
}
}
func TestAgentHeartbeatVerdictNeverSeenIsOffline(t *testing.T) {
now := time.Unix(1_700_000_000, 0)
patch, online := agentHeartbeatVerdict(0, now)
if online {
t.Fatal("never-seen agent must read offline")
}
if patch.LatencyMs != -1 || patch.LastHeartbeat != 0 || patch.LastError != "" {
t.Fatalf("patch = %+v, want zero stamp, latency -1, no error", patch)
}
}
+27 -2
View File
@@ -164,13 +164,38 @@ func (s *PanelService) GetUpdateInfo() (*PanelUpdateInfo, error) {
// display version. An unparseable tag (not a V-series release) yields ""
// so the UI falls back to the raw fields.
func displayUpdateFields(latestTag string) (string, bool) {
return displayUpdateFieldsFor(latestTag, config.GetMasterDisplayVersion(), config.GetBaseVersion())
}
// displayUpdateFieldsFor is displayUpdateFields with the running
// build's versions as inputs. When the DISPLAY comparison cannot
// judge — the running build's display version is itself unparseable,
// the "VV2.1.0" double-letter builds — the verdict falls back to the
// raw version comparison instead of reporting "no update": the
// display path used to swallow a real update entirely (the panel
// showed no prompt while /releases/latest happily served the newer
// tag, and the native updater refused to run). A parseable display
// version keeps the segment rule exactly as before.
func displayUpdateFieldsFor(latestTag, currentDisplay, currentRaw string) (string, bool) {
parts, ok := config.ParseDisplayVersion(latestTag)
if !ok {
return "", false
}
disp := fmt.Sprintf("V%d.%d.%d", parts[0], parts[1], parts[2])
cmp, ok := config.CompareDisplayVersion(disp, config.GetMasterDisplayVersion())
return disp, ok && cmp > 0
cmp, ok := config.CompareDisplayVersion(disp, currentDisplay)
if !ok {
// Raw fallback: judge the latest tag against the running
// build's raw base version, parsed with the same tolerant
// display parser (bare and V/A forms both parse). An
// unparseable raw base yields no verdict → false. This is
// deliberately NOT isNewerVersion: on this fork's uppercase-V
// tags that helper degrades to string inequality (its
// normalizer strips only a lowercase "v"), which would report
// a phantom update whenever latest merely differs in form.
rawCmp, rawOk := config.CompareDisplayVersion(disp, currentRaw)
return disp, rawOk && rawCmp > 0
}
return disp, cmp > 0
}
// devChannelActive reports whether self-update should track the rolling dev
@@ -25,6 +25,42 @@ func TestDisplayUpdateFields(t *testing.T) {
}
}
// V2.2.1 ③: when the running build's display version is unparseable
// (the "VV2.1.0" double-letter builds), the display comparison cannot
// judge and the verdict must fall back to the raw version comparison
// instead of swallowing the update. A parseable display version keeps
// the segment rule — the raw string plays no part there.
func TestDisplayUpdateFieldsForFallback(t *testing.T) {
// Unparseable display "VV2.1.0", newer latest → available via raw.
if disp, avail := displayUpdateFieldsFor("V2.2.0", "VV2.1.0", "2.1.0"); disp != "V2.2.0" || !avail {
t.Fatalf("VV display, newer latest = %q, %v; want V2.2.0, true", disp, avail)
}
// Unparseable display, latest equal to the raw base → not
// available, in both raw forms (bare and V-prefixed).
if disp, avail := displayUpdateFieldsFor("V2.1.0", "VV2.1.0", "2.1.0"); disp != "V2.1.0" || avail {
t.Fatalf("VV display, same latest (bare raw) = %q, %v; want V2.1.0, false", disp, avail)
}
if _, avail := displayUpdateFieldsFor("V2.1.0", "VV2.1.0", "V2.1.0"); avail {
t.Fatal("VV display, same latest (V raw) must not report an update")
}
// Unparseable display AND unparseable raw base → no verdict, false.
if _, avail := displayUpdateFieldsFor("V2.2.0", "VV2.1.0", "3.7.0-agent.1"); avail {
t.Fatal("unparseable raw base must yield no update verdict")
}
// Parseable display: segment rule decides, raw base is ignored
// even when it disagrees.
if disp, avail := displayUpdateFieldsFor("V2.2.0", "V2.1.0", "9.9.9"); disp != "V2.2.0" || !avail {
t.Fatalf("parseable display, newer = %q, %v; want V2.2.0, true", disp, avail)
}
if _, avail := displayUpdateFieldsFor("V2.1.0", "V2.1.0", "1.0.0"); avail {
t.Fatal("parseable display, equal versions must not report an update (raw base must not leak in)")
}
// Unparseable latest tag still yields "" so the UI uses raw fields.
if disp, avail := displayUpdateFieldsFor("3.7.0-agent.1", "VV2.1.0", "2.1.0"); disp != "" || avail {
t.Fatalf("legacy tag = %q, %v; want empty, false", disp, avail)
}
}
func TestServiceUnitFromCgroup(t *testing.T) {
content := "0::/system.slice/x-ui.service\n"
if got := serviceUnitFromCgroup(content); got != "x-ui" {
+27
View File
@@ -34,6 +34,9 @@ type Meta struct {
// corrections in GB (V2.0.1), either sign: added to the counted
// uplink (tx) / downlink (rx) figures wherever the ledger is
// read — panel statistics, the public page and traffic alerts.
// The stored values are the STANDING accumulated correction;
// the edit form submits one-shot deltas that MergeCalDelta folds
// into them (V2.2.1 ⑥b).
TrafficCalUpGB float64
TrafficCalDownGB float64
// StatIface pins the one interface whose counters the agent
@@ -138,6 +141,30 @@ func (s *Store) UpsertMeta(m Meta) error {
return err
}
// MergeCalDelta folds a probe-meta form submission into the stored
// row under the one-shot calibration semantics (V2.2.1 ⑥b): the
// form's calibration fields are a correction TO APPLY NOW, not the
// new standing value. A non-zero submitted value adds onto the stored
// standing correction; a zero field (an absent field is the same
// zero — the wire type cannot distinguish them, and both mean
// "nothing to apply") keeps the stored value untouched. Every other
// field takes the submitted value, as UpsertMeta's full replace
// does. Paired with the form clearing its calibration inputs after a
// successful save, a later unrelated save can never re-apply an old
// correction.
func MergeCalDelta(existing, submitted Meta) Meta {
out := submitted
out.TrafficCalUpGB = existing.TrafficCalUpGB
if submitted.TrafficCalUpGB != 0 {
out.TrafficCalUpGB += submitted.TrafficCalUpGB
}
out.TrafficCalDownGB = existing.TrafficCalDownGB
if submitted.TrafficCalDownGB != 0 {
out.TrafficCalDownGB += submitted.TrafficCalDownGB
}
return out
}
func boolToInt(b bool) int64 {
if b {
return 1
@@ -0,0 +1,143 @@
package probe
import (
"path/filepath"
"testing"
"time"
)
// V2.2.1 ⑥: GB entries convert on the binary convention (1024^3)
// every formatter uses, stored decimal-era limits are rescaled once,
// and calibration submissions are one-shot deltas folded into the
// standing correction — never re-applied by a later unrelated save.
func TestCalBytesBinaryConvention(t *testing.T) {
if got := calBytes(3000); got != 3000*(1<<30) {
t.Fatalf("calBytes(3000) = %d, want %d (3000 GiB)", got, 3000*(1<<30))
}
if got := calBytes(0); got != 0 {
t.Fatalf("calBytes(0) = %d, want 0", got)
}
if got := calBytes(-1.5); got != -1610612736 {
t.Fatalf("calBytes(-1.5) = %d, want -1610612736", got)
}
}
func TestGBConventionMigrationRescalesLimitsOnce(t *testing.T) {
path := filepath.Join(t.TempDir(), "probe.db")
st, err := Open(path)
if err != nil {
t.Fatalf("open: %v", err)
}
// Seed a legacy row as a pre-V2.2.1 database holds it — 3000 GB
// typed at the decimal convention = 3.0e12 stored bytes — and
// strike the migration mark so the next open sees an unmigrated
// database.
if _, err := st.db.Exec(`INSERT INTO probe_meta (agent_id, traffic_limit) VALUES (9, 3000000000000)`); err != nil {
t.Fatalf("seed legacy limit: %v", err)
}
if _, err := st.db.Exec(`DELETE FROM probe_setting WHERE key = ?`, gbConventionKey); err != nil {
t.Fatalf("strike mark: %v", err)
}
if err := st.Close(); err != nil {
t.Fatalf("close: %v", err)
}
st2, err := Open(path)
if err != nil {
t.Fatalf("reopen (migration run): %v", err)
}
m, found, err := st2.Meta(9)
if err != nil || !found {
t.Fatalf("meta after migration: found=%v err=%v", found, err)
}
// 3.0e12 x 1024^3/1e9 = 3000 x 2^30 exactly (~3.221e12).
if want := int64(3000) * (1 << 30); m.TrafficLimit != want {
t.Fatalf("migrated limit = %d, want %d", m.TrafficLimit, want)
}
if err := st2.Close(); err != nil {
t.Fatalf("close 2: %v", err)
}
st3, err := Open(path)
if err != nil {
t.Fatalf("reopen 2: %v", err)
}
defer st3.Close()
m, found, err = st3.Meta(9)
if err != nil || !found {
t.Fatalf("meta after second open: found=%v err=%v", found, err)
}
if want := int64(3000) * (1 << 30); m.TrafficLimit != want {
t.Fatalf("limit after second open = %d, want %d (no double rescale)", m.TrafficLimit, want)
}
}
func TestOneShotCalibrationSaveFlow(t *testing.T) {
st := newTestStore(t)
now := time.Now()
book(t, st, 5, "boot-a", 1000, 1000, now, nil, 0) // baseline: books 0
book(t, st, 5, "boot-a", 3000, 2000, now, nil, 0) // books rx 2000, tx 1000
// First save carries a +1 GB uplink correction delta.
submitted := DefaultMeta(5)
submitted.TrafficCalUpGB = 1
merged := MergeCalDelta(DefaultMeta(5), submitted)
if merged.TrafficCalUpGB != 1 {
t.Fatalf("standing after first fold = %v, want 1", merged.TrafficCalUpGB)
}
if err := st.UpsertMeta(merged); err != nil {
t.Fatalf("upsert 1: %v", err)
}
after1, ok, err := st.TrafficFor(5)
if err != nil || !ok {
t.Fatalf("traffic after save 1: ok=%v err=%v", ok, err)
}
if want := int64(1000) + (1 << 30); after1.TotalTx != want {
t.Fatalf("TotalTx after save 1 = %d, want %d", after1.TotalTx, want)
}
if after1.TotalRx != 2000 {
t.Fatalf("TotalRx after save 1 = %d, want 2000 (uncalibrated)", after1.TotalRx)
}
// Second save changes another field; the (cleared) form submits
// zero calibration — the calibrated figure must not move again.
stored, found, err := st.Meta(5)
if err != nil || !found {
t.Fatalf("stored meta: found=%v err=%v", found, err)
}
sub2 := DefaultMeta(5)
sub2.GroupName = "g2"
merged2 := MergeCalDelta(stored, sub2)
if merged2.TrafficCalUpGB != 1 {
t.Fatalf("standing after zero-delta save = %v, want 1 (unchanged)", merged2.TrafficCalUpGB)
}
if err := st.UpsertMeta(merged2); err != nil {
t.Fatalf("upsert 2: %v", err)
}
after2, ok, err := st.TrafficFor(5)
if err != nil || !ok {
t.Fatalf("traffic after save 2: ok=%v err=%v", ok, err)
}
if after2.TotalTx != after1.TotalTx {
t.Fatalf("TotalTx moved on an unrelated save: %d -> %d (double-applied calibration)", after1.TotalTx, after2.TotalTx)
}
// A further non-zero delta accumulates exactly once more.
sub3 := DefaultMeta(5)
sub3.TrafficCalUpGB = 0.5
merged3 := MergeCalDelta(merged2, sub3)
if merged3.TrafficCalUpGB != 1.5 {
t.Fatalf("standing after second delta = %v, want 1.5", merged3.TrafficCalUpGB)
}
if err := st.UpsertMeta(merged3); err != nil {
t.Fatalf("upsert 3: %v", err)
}
after3, _, err := st.TrafficFor(5)
if err != nil {
t.Fatalf("traffic after save 3: %v", err)
}
if want := int64(1000) + calBytes(1.5); after3.TotalTx != want {
t.Fatalf("TotalTx after save 3 = %d, want %d", after3.TotalTx, want)
}
}
+4 -2
View File
@@ -91,8 +91,10 @@ func TestTrafficCalibrationAppliedAndFloored(t *testing.T) {
t.Fatalf("AllTraffic: %v", err)
}
tr := all[5]
if tr.TotalRx != 3_500_000_000 || tr.MonthRx != 3_500_000_000 || tr.DayRx != 3_500_000_000 {
t.Errorf("rx figures after +0.5 GB calibration = %d/%d/%d, want 3500000000 each", tr.TotalRx, tr.MonthRx, tr.DayRx)
// V2.2.1 ⑥: calibration converts on the binary convention —
// +0.5 GB = 0.5 x 2^30 = 536870912 B (was 5e8 at the old 1e9).
if tr.TotalRx != 3_536_870_912 || tr.MonthRx != 3_536_870_912 || tr.DayRx != 3_536_870_912 {
t.Errorf("rx figures after +0.5 GB calibration = %d/%d/%d, want 3536870912 each", tr.TotalRx, tr.MonthRx, tr.DayRx)
}
if tr.TotalTx != 0 || tr.MonthTx != 0 || tr.DayTx != 0 {
t.Errorf("tx figures after -2 GB calibration = %d/%d/%d, want 0 (floored)", tr.TotalTx, tr.MonthTx, tr.DayTx)
+43
View File
@@ -213,9 +213,52 @@ func Open(path string) (*Store, error) {
db.Close()
return nil, fmt.Errorf("probe schema migration: %w", err)
}
if err := migrateGBConvention(db); err != nil {
db.Close()
return nil, fmt.Errorf("probe gb migration: %w", err)
}
return &Store{db: db, nodes: make(map[int64]*nodeState)}, nil
}
// gbConventionKey is the probe_setting mark recording that the
// V2.2.1 ⑥ traffic-limit rescale has been applied to this database.
const gbConventionKey = "gb_1024_migrated_v1"
// migrateGBConvention rescales stored probe_meta traffic limits from
// the old decimal entry convention to the unified binary one, once.
// Limits are typed in GB and were stored as GB × 1e9 bytes, while
// every byte formatter (panel, public page) divides by 1024^3 — a
// 3000 GB limit displayed as 2.73 TB instead of 2.93 TB. The GB the
// operator typed is the intent, so stored bytes are scaled by
// 1024^3/1e9 = 1.073741824 and now read back correctly under the
// binary conversion. Only traffic_limit is byte-denominated:
// calibration values are stored in GB (their byte conversion,
// calBytes, carries the fix) and price/sort/reset-day are unitless —
// neither is touched here. The probe_setting mark makes the rescale
// run exactly once per database.
func migrateGBConvention(db *sql.DB) error {
var v string
err := db.QueryRow(`SELECT value FROM probe_setting WHERE key = ?`, gbConventionKey).Scan(&v)
if err == nil {
return nil // already applied
}
if err != sql.ErrNoRows {
return err
}
tx, err := db.Begin()
if err != nil {
return err
}
defer tx.Rollback()
if _, err := tx.Exec(`UPDATE probe_meta SET traffic_limit = CAST(ROUND(traffic_limit * 1.073741824) AS INTEGER) WHERE traffic_limit > 0`); err != nil {
return err
}
if _, err := tx.Exec(`INSERT OR REPLACE INTO probe_setting (key, value) VALUES (?, '1')`, gbConventionKey); err != nil {
return err
}
return tx.Commit()
}
// metaColumnMigrations are the probe_meta columns added after the
// table's first release; CREATE TABLE IF NOT EXISTS never retrofits
// them onto an existing database, so they are added here, once, by
+6 -2
View File
@@ -332,9 +332,13 @@ func (s *Store) TrafficFor(agentID int64) (Traffic, bool, error) {
// AllTraffic returns every agent's ledger in one query, gated as in
// TrafficFor — the bulk read port for B3's node list.
// calBytes converts a GB calibration value to bytes on the panel's
// 1 GB = 1e9 B convention (the same one the traffic limit uses).
// binary convention, 1 GB = 1024^3 B — the convention every byte
// formatter in the panel and on the public page uses, and the one
// the traffic limit's GB entry converts with (V2.2.1 ⑥; the decimal
// 1e9 stood here before, which made a 3000 GB limit read as 2.73 TB
// instead of 2.93 TB wherever bytes were formatted).
func calBytes(gb float64) int64 {
return int64(math.Round(gb * 1e9))
return int64(math.Round(gb * (1 << 30)))
}
// applyCalibration shifts a ledger row by the operator's per-node
+14 -1
View File
@@ -268,6 +268,19 @@ func expiresInDays(expires *string, now time.Time) any {
return days
}
// probeDisplayName is the name a node carries in the public views
// (V2.2.1 ⑦): the operator's public remark when one is set — it is
// the label written for public consumption — else the node name. The
// contract field stays "name"; themes and the public API keep their
// shape, only the value's source changes. The panel's internal node
// list is a different surface and keeps the plain node name.
func probeDisplayName(name, publicRemark string) string {
if r := strings.TrimSpace(publicRemark); r != "" {
return r
}
return name
}
// buildProbeNodeView renders one node for one audience. The public
// shape is built by whitelist (hub node_view): address, hostname and
// the private remark exist only in the admin (full) rendering.
@@ -298,7 +311,7 @@ func buildProbeNodeView(src probeNodeSource, full bool) map[string]any {
view := map[string]any{
"id": a.Id,
"name": src.name,
"name": probeDisplayName(src.name, m.PublicRemark),
"country": m.CountryPin,
"group": m.GroupName,
"public_remark": m.PublicRemark,
@@ -0,0 +1,40 @@
package service
import (
"testing"
"github.com/mhsanaei/3x-ui/v3/internal/web/service/probe"
)
// V2.2.1 ⑦: the public views' node name is the operator's public
// remark when one is set, and the node name otherwise — the panel's
// own list (a different surface) is untouched.
func TestProbeDisplayNamePrefersPublicRemark(t *testing.T) {
if got := probeDisplayName("node-name", "东京 01"); got != "东京 01" {
t.Fatalf("with remark = %q, want the remark", got)
}
if got := probeDisplayName("node-name", ""); got != "node-name" {
t.Fatalf("empty remark = %q, want node name", got)
}
if got := probeDisplayName("node-name", " "); got != "node-name" {
t.Fatalf("whitespace-only remark = %q, want node name", got)
}
}
func TestBuildProbeNodeViewNameSource(t *testing.T) {
withRemark := probeNodeSource{
name: "internal-name",
meta: probe.Meta{PublicRemark: "公开名", Public: true},
}
if got := buildProbeNodeView(withRemark, false)["name"]; got != "公开名" {
t.Fatalf("anonymous view name = %v, want the public remark", got)
}
if got := buildProbeNodeView(withRemark, true)["name"]; got != "公开名" {
t.Fatalf("admin view name = %v, want the public remark", got)
}
without := probeNodeSource{name: "internal-name", meta: probe.Meta{Public: true}}
if got := buildProbeNodeView(without, false)["name"]; got != "internal-name" {
t.Fatalf("view name without remark = %v, want the node name", got)
}
}