AGC 开发态补齐后台 Web 启动与端口汇总,端口归属探测从 WMI 换成 netstat

- dev-port.mjs 新增 resolveAgcAdminWebEndpoint:后台 Web 端口按 Linux 用户端口段 start + 3、非 Linux 优先 3102 解析,ADMIN_WEB_PORT 可显式指定,并把已解析的 AGC Vite 端口列为保留端口
- start-dev-stack.mjs 随 AGC 一起拉起 apps/admin-web(AGC_DEV_ADMIN_WEB=0 关闭):与 AGC Vite 一样由启动器直接持有并纳入信号与退出收束,不走会整体重写 .app/dev-stack.json 的 dev:admin-web
- start-dev-stack.mjs 新增启动汇总行,一次给出前端、后端、后台、数据库与 bgfilter-worker 的实际地址,端口漂移后以该行为准
- start-dev-stack.mjs 端口归属探测改用 netstat -ano 取端口到 PID、.NET Process 读可执行文件路径,仅在核对 SpacetimeDB --data-dir 归属时按 PID 取命令行并做 5 分钟 TTL 缓存:本机实测单轮探测由约 43 秒降到 0.37 秒,agc:serve 的 starting backend stack 到 backend ready 由约 80 秒降到 16.7 秒
- 后台 Web 端口解析或启动失败只告警,不阻断也不连带停止 AGC 客户端与配套后端
- tests/dev-port.test.ts、tests/start-dev-stack.test.ts 补充用例覆盖后台 Web 端口解析与严格占用、失败软化、启动汇总、netstat 探测脚本与命令行缓存失效
- docs/【开发运维】与 docs/technical 同步 AGC 开发态启动口径,pitfalls.md 记录 WMI 慢速探测的实测数据、netstat 改法与 PID 取最后一列的易错点
This commit is contained in:
2026-09-14 20:55:51 +08:00
parent 330e4d6c5b
commit 2bdee990a2
7 changed files with 672 additions and 21 deletions
@@ -14,12 +14,14 @@ import {
import {
agcVitePortEnvKey,
readAgcDevEndpoint,
resolveAgcAdminWebEndpoint,
resolveAgcDevEndpoint,
withAgcDevEndpointEnv,
} from './dev-port.mjs';
const appRoot = fileURLToPath(new URL('..', import.meta.url));
const repoRoot = resolve(appRoot, '../..');
const adminWebDir = resolve(repoRoot, 'apps/admin-web');
const devStackStatePath = resolve(repoRoot, '.app/dev-stack.json');
const apiServerExePath = resolve(
repoRoot,
@@ -32,6 +34,8 @@ const backendSpacetimeDataDir = resolve(
repoRoot,
'server-rs/.spacetimedb/ai-game-creator/data',
);
// 后台 Web 默认跟随 AGC 一起起来,便于联调后台页面;`AGC_DEV_ADMIN_WEB=0` 可关闭。
const agcDevAdminWebEnvKey = 'AGC_DEV_ADMIN_WEB';
const npm = process.platform === 'win32' ? 'npm.cmd' : 'npm';
const childLifecycles = new WeakMap();
@@ -200,36 +204,122 @@ function urlPort(url) {
}
}
// 读取端口当前真正的监听进程身份。返回 null 表示探测本身不可用(例如缺少
// Get-NetTCPConnection),此时调用方必须退化为旧行为,不能让本地启动直接失败。
// 端口归属探测脚本。历史实现用 `Get-NetTCPConnection` 取监听进程,而它底层走
// WMI:实测单端口单次 11.2 秒、再叠加每个 PID 的 `Get-CimInstance` 3.3 秒,
// 一轮探测约 43 秒,直接把"配套后端就绪"等待拖到分钟级。改用原生
// `netstat -ano`(约 30 毫秒)取端口 -> PID,再用 .NET `Process` 读进程名和
// 可执行文件路径(毫秒级);只有核对 SpacetimeDB `--data-dir` 归属时才按 PID
// 取命令行,并允许调用方把已知命令行传进来复用。
const windowsPortOwnerProbeCommand = [
'$ErrorActionPreference = "SilentlyContinue"',
'$queriedPorts = @()',
'foreach ($raw in ($env:GENARRATIVE_QUERY_PORTS -split ",")) {',
' if ($raw -match "^\\d+$") { $queriedPorts += [int]$raw }',
'}',
'$knownCommandLines = @{}',
'if ($env:GENARRATIVE_KNOWN_COMMAND_LINES) {',
' try {',
' foreach ($property in (ConvertFrom-Json $env:GENARRATIVE_KNOWN_COMMAND_LINES).PSObject.Properties) {',
' $knownCommandLines[[int]$property.Name] = [string]$property.Value',
' }',
' } catch { }',
'}',
'$listenerPidByPort = @{}',
'foreach ($line in (netstat -ano -p tcp)) {',
' $fields = @($line -split "\\s+" | Where-Object { $_ })',
' if ($fields.Count -lt 4) { continue }',
' if ($fields[0] -ne "TCP") { continue }',
' # A listening socket always has foreign address 0.0.0.0:0 / [::]:0, which',
' # is locale-independent unlike the localized netstat State column.',
' if ($fields[2] -notmatch ":0$") { continue }',
' $localPort = [int]($fields[1].Split(":")[-1])',
' if ($queriedPorts -notcontains $localPort) { continue }',
' # The PID is the last column; do not hardcode its index.',
' if ($fields[-1] -notmatch "^\\d+$") { continue }',
' $listenerPidByPort[$localPort] = [int]$fields[-1]',
'}',
'$result = @()',
'foreach ($port in ($listenerPidByPort.Keys | Sort-Object)) {',
' $processId = $listenerPidByPort[$port]',
' $name = $null',
' $executablePath = $null',
' $commandLine = $null',
' try {',
' $process = [System.Diagnostics.Process]::GetProcessById($processId)',
' $name = $process.ProcessName + ".exe"',
' try { $executablePath = $process.MainModule.FileName } catch { }',
' } catch { }',
' if ($knownCommandLines.ContainsKey($processId)) {',
' $commandLine = $knownCommandLines[$processId]',
' } elseif (($name -like "spacetime*") -or (-not $executablePath)) {',
' try { $commandLine = (Get-CimInstance Win32_Process -Filter ("ProcessId=" + $processId)).CommandLine } catch { }',
' }',
' $result += [pscustomobject]@{ port = [int]$port; processId = $processId; name = $name; executablePath = $executablePath; commandLine = $commandLine }',
'}',
'ConvertTo-Json -InputObject @($result) -Compress',
].join('\n');
// 进程命令行在进程生命周期内不变,但 PID 会被系统复用;按 PID 记 TTL 缓存,
// 让"等配套后端就绪"的轮询只在首个周期付出 WMI 成本。TTL 取 5 分钟:本轮实测
// 这台机器上首次 WMI 调用约 18 秒(热调用 3.3 秒),而 PID 在 5 分钟内被复用
// 成另一个运行本工作树 data dir 的 SpacetimeDB 才能造成误判,概率可忽略。
// 默认实现才缓存,注入实现(测试)与显式 env 始终重新读取。
const WINDOWS_COMMAND_LINE_CACHE_TTL_MS = 300_000;
const windowsPortOwnerCommandLineCache = new Map();
function resolveCommandLineCache({ spawnImpl, env }) {
return spawnImpl === spawnSync && env === process.env
? windowsPortOwnerCommandLineCache
: new Map();
}
// 读取端口当前真正的监听进程身份。返回 null 表示探测本身不可用(例如系统缺少
// netstat),此时调用方必须退化为旧行为,不能让本地启动直接失败。
function readWindowsPortOwnerIdentities(
ports,
{ spawnImpl = spawnSync, env = process.env } = {},
{
spawnImpl = spawnSync,
env = process.env,
now = Date.now,
commandLineTtlMs = WINDOWS_COMMAND_LINE_CACHE_TTL_MS,
commandLineCache = resolveCommandLineCache({ spawnImpl, env }),
} = {},
) {
const uniquePorts = [...new Set(ports.filter((port) => port > 0))];
if (uniquePorts.length === 0) {
return null;
}
const command = [
'$ErrorActionPreference = "SilentlyContinue"',
'$ports = ($env:GENARRATIVE_QUERY_PORTS -split ",") | Where-Object { $_ }',
'$result = @()',
'foreach ($port in $ports) {',
' $connection = Get-NetTCPConnection -State Listen -LocalPort ([int]$port) -ErrorAction SilentlyContinue | Select-Object -First 1',
' if (-not $connection) { continue }',
' $owner = Get-CimInstance Win32_Process -Filter ("ProcessId=" + $connection.OwningProcess) -ErrorAction SilentlyContinue',
' $result += [pscustomobject]@{ port = [int]$port; processId = [int]$connection.OwningProcess; name = $owner.Name; executablePath = $owner.ExecutablePath; commandLine = $owner.CommandLine }',
'}',
'ConvertTo-Json -InputObject @($result) -Compress',
].join('\n');
const knownCommandLines = {};
for (const [processId, record] of [...commandLineCache]) {
if (record && now() - record.at < commandLineTtlMs) {
knownCommandLines[processId] = record.commandLine;
} else {
commandLineCache.delete(processId);
}
}
const childEnv = {
...env,
GENARRATIVE_QUERY_PORTS: uniquePorts.join(','),
};
if (Object.keys(knownCommandLines).length > 0) {
childEnv.GENARRATIVE_KNOWN_COMMAND_LINES =
JSON.stringify(knownCommandLines);
}
const result = spawnImpl(
'powershell.exe',
['-NoProfile', '-ExecutionPolicy', 'Bypass', '-Command', command],
[
'-NoProfile',
'-ExecutionPolicy',
'Bypass',
'-Command',
windowsPortOwnerProbeCommand,
],
{
encoding: 'utf8',
env: { ...env, GENARRATIVE_QUERY_PORTS: uniquePorts.join(',') },
env: childEnv,
maxBuffer: 8 * 1024 * 1024,
},
);
@@ -240,9 +330,22 @@ function readWindowsPortOwnerIdentities(
const owners = new Map();
for (const entry of parseWindowsProcessSnapshot(result.stdout)) {
const port = Number(entry?.port);
if (Number.isInteger(port) && port > 0) {
owners.set(port, entry);
if (!Number.isInteger(port) || port <= 0) {
continue;
}
const processId = Number(entry?.processId);
if (
Number.isInteger(processId) &&
processId > 0 &&
typeof entry?.commandLine === 'string' &&
entry.commandLine
) {
commandLineCache.set(processId, {
commandLine: entry.commandLine,
at: now(),
});
}
owners.set(port, entry);
}
return owners;
}
@@ -879,10 +982,102 @@ async function startVite(apiTarget, endpoint = readAgcDevEndpoint()) {
);
}
function readAdminWebEnabled(env = process.env) {
return String(env[agcDevAdminWebEnvKey] ?? '').trim() !== '0';
}
// 后台 Web 与 AGC Vite 一样直接由本启动器持有,不经过 `dev.mjs admin-web`:
// 后者会整体重写 `.app/dev-stack.json`,把本次配套后端的状态覆盖掉。
function startAdminWeb(
apiUrl,
endpoint,
{ env = process.env, spawnImpl = spawnChild } = {},
) {
return spawnImpl(
npm,
[
'--prefix',
'../..',
'exec',
'vite',
'--',
'--host',
endpoint.host,
'--port',
String(endpoint.port),
'--strictPort',
],
{
cwd: adminWebDir,
env: {
...env,
ADMIN_API_TARGET: apiUrl,
GENARRATIVE_API_TARGET: apiUrl,
GENARRATIVE_API_PORT: String(urlPort(apiUrl) || 8082),
ADMIN_WEB_BASE: endpoint.basePath,
},
},
);
}
function formatStartupSummary({
frontendUrl = '',
apiUrl = '',
adminWebUrl = '',
spacetimeUrl = '',
bgfilterWorkerUrl = '',
} = {}) {
const segments = [
['前端', frontendUrl],
['后端', apiUrl],
['后台', adminWebUrl],
['数据库', spacetimeUrl],
['bgfilter-worker', bgfilterWorkerUrl],
]
.filter(([, value]) => Boolean(value))
.map(([label, value]) => `${label} ${value}`);
return `[ai-game-creator-shell] 启动汇总: ${segments.join(' | ')}`;
}
// 后台 Web 是可选联调服务:端口解析或启动失败只告警,不能阻断 AGC 客户端与配套后端。
async function ensureAdminWeb({
apiUrl,
reservedPorts = [],
env = process.env,
enabled = readAdminWebEnabled(env),
resolveEndpoint = resolveAgcAdminWebEndpoint,
spawnAdminWeb = startAdminWeb,
waitForExit = waitForChildTermination,
warn = (message) => console.warn(message),
} = {}) {
if (!enabled) {
return { endpoint: null, child: null };
}
try {
const endpoint = await resolveEndpoint({ env, reservedPorts });
const child = spawnAdminWeb(apiUrl, endpoint, { env });
waitForExit(child).then((failure) => {
warn(
`[ai-game-creator-shell] 后台 Web 已退出(${formatChildFailure(failure)}),AGC 继续运行。`,
);
});
return { endpoint, child };
} catch (error) {
warn(
`[ai-game-creator-shell] 后台 Web 未能启动(${
error instanceof Error ? error.message : String(error)
}),AGC 继续运行。`,
);
return { endpoint: null, child: null };
}
}
async function main() {
let backendChild = null;
let startedBackend = false;
let viteChild = null;
let adminWebChild = null;
let shutdownSignal = '';
const signalHandlers = new Map();
@@ -905,6 +1100,7 @@ async function main() {
const handler = () => {
shutdownSignal = signal;
stopChild(viteChild, signal);
stopChild(adminWebChild, signal);
stopChild(backendChild, signal);
// 立刻清扫,避免外层 taskkill /F 抢在 finally 之前把本进程杀掉。
sweepStartedBackend();
@@ -937,6 +1133,25 @@ async function main() {
throw new Error(`启动期收到 ${shutdownSignal},已停止前端服务`);
}
const adminWeb = await ensureAdminWeb({
apiUrl: backend.targets.apiUrl,
// AGC Vite 端口尚未监听,必须显式保留,避免被后台 Web 抢先占用。
reservedPorts: [endpoint.port],
});
adminWebChild = adminWeb.child;
if (shutdownSignal) {
throw new Error(`启动期收到 ${shutdownSignal},已停止后台 Web`);
}
console.log(
formatStartupSummary({
frontendUrl: endpoint.url,
apiUrl: backend.targets.apiUrl,
adminWebUrl: adminWeb.endpoint?.url ?? '',
spacetimeUrl: backend.targets.spacetimeUrl,
bgfilterWorkerUrl: backend.targets.bgfilterWorkerUrl,
}),
);
const children = [backendChild, viteChild].filter(Boolean);
if (children.length === 0) {
return 0;
@@ -946,10 +1161,12 @@ async function main() {
children.map((child) => waitForChildTermination(child)),
);
stopChild(viteChild);
stopChild(adminWebChild);
stopChild(backendChild);
return failure.type === 'error' || failure.signal ? 1 : (failure.code ?? 0);
} catch (error) {
stopChild(viteChild);
stopChild(adminWebChild);
stopChild(backendChild);
console.error(
`[ai-game-creator-shell] ${error instanceof Error ? error.message : String(error)}`,
@@ -958,6 +1175,7 @@ async function main() {
} finally {
await Promise.all([
terminateChildTree(viteChild),
terminateChildTree(adminWebChild),
terminateChildTree(backendChild),
]);
sweepStartedBackend();
@@ -975,9 +1193,12 @@ function isDirectModuleExecution() {
}
export {
agcDevAdminWebEnvKey,
ensureAdminWeb,
ensureBackend,
formatChildFailure,
formatOwnerLabel,
formatStartupSummary,
isAiGameCreatorServer,
isBackendReady,
isDirectModuleExecution,
@@ -985,6 +1206,7 @@ export {
isWorktreeApiServerOwner,
isWorktreeSpacetimeOwner,
preflightExistingVite,
readAdminWebEnabled,
readBackendServiceFailure,
readChildFailure,
readExistingViteServer,
@@ -993,6 +1215,7 @@ export {
resolveBackendTargetsFromState,
runWindowsTaskkill,
spawnChild,
startAdminWeb,
stopChild,
terminateChildTree,
verifyAgcBackendOwnership,