server.js v3.1: 新增GPU/训练/Agent API端点 + 算力池API + 综合状态端点
This commit is contained in:
bingshuo
parent
831f462fae
commit
8e07caf2d1
@ -1,5 +1,5 @@
|
||||
// # 光湖服务器主控台 API
|
||||
// # Guanghu Server Console v2.0 — with MCP Terminal
|
||||
// # Guanghu Server Console v3.1 — with GPU & Agent API
|
||||
// # 部署: guanghulab.com/console/
|
||||
// # 版权: 国作登字-2026-A-00037559 · TCS-0002∞
|
||||
|
||||
@ -13,6 +13,12 @@ const app = express();
|
||||
app.use(express.json());
|
||||
app.use(express.static(path.join(__dirname, 'public')));
|
||||
|
||||
// 数据存储目录
|
||||
const DATA_DIR = '/opt/zhuyuan/data';
|
||||
|
||||
// 确保数据目录存在
|
||||
try { fs.mkdirSync(DATA_DIR, { recursive: true }); } catch(e) {}
|
||||
|
||||
// 六台服务器配置（从 gatekeeper-deployment.json 读取）
|
||||
const SERVERS = [
|
||||
{ code: "BS-GZ-006", name: "广州 · 代码仓库", ip: "43.139.217.141", key: "zy_gtw_4a155446b7acc09fe46a2a29972c0ca5d9954da19c35dc23", port: 3910 },
|
||||
@ -66,6 +72,21 @@ async function pingServer(srv) {
|
||||
}
|
||||
}
|
||||
|
||||
// ========== Agent认证 ==========
|
||||
|
||||
function checkAgentAuth(req) {
|
||||
const auth = req.headers.authorization || '';
|
||||
const token = auth.replace('Bearer ', '');
|
||||
if (!token) return false;
|
||||
try {
|
||||
const keysData = fs.readFileSync('/tmp/zhuyuan-keys.json', 'utf-8');
|
||||
const keys = JSON.parse(keysData || '[]');
|
||||
return keys.some(k => k.value === token);
|
||||
} catch(e) {
|
||||
return false;
|
||||
}
|
||||
}
|
||||
|
||||
// ========== API 路由 ==========
|
||||
|
||||
// 1. 获取所有服务器状态
|
||||
@ -355,10 +376,305 @@ app.get('/api/terminal/sessions', (req, res) => {
|
||||
res.json({ ok: true, sessions });
|
||||
});
|
||||
|
||||
// ========== 密钥投递 API ==========
|
||||
|
||||
// 密钥存储
|
||||
const KEYS_FILE = '/tmp/zhuyuan-keys.json';
|
||||
let keyStore = [];
|
||||
|
||||
function loadKeys() {
|
||||
try {
|
||||
if (fs.existsSync(KEYS_FILE)) {
|
||||
const data = fs.readFileSync(KEYS_FILE, 'utf-8');
|
||||
keyStore = JSON.parse(data || '[]');
|
||||
}
|
||||
} catch(e) { keyStore = []; }
|
||||
}
|
||||
function saveKeys() {
|
||||
fs.writeFileSync(KEYS_FILE, JSON.stringify(keyStore, null, 2));
|
||||
}
|
||||
loadKeys();
|
||||
|
||||
// 14. 投递密钥
|
||||
app.post('/api/keys/submit', (req, res) => {
|
||||
const { value, label } = req.body;
|
||||
if (!value) return res.status(400).json({ error: '缺少密钥内容' });
|
||||
const id = 'key_' + crypto.randomBytes(6).toString('hex');
|
||||
keyStore.push({ id, value, label: label || '未命名', created_at: new Date().toISOString() });
|
||||
saveKeys();
|
||||
res.json({ ok: true, id });
|
||||
});
|
||||
|
||||
// 15. 列出所有密钥（不包含值）
|
||||
app.get('/api/keys', (req, res) => {
|
||||
const list = keyStore.map(k => ({ id: k.id, label: k.label, created_at: k.created_at }));
|
||||
res.json({ ok: true, keys: list });
|
||||
});
|
||||
|
||||
// 16. 获取单个密钥值
|
||||
app.get('/api/keys/retrieve/:id', (req, res) => {
|
||||
const key = keyStore.find(k => k.id === req.params.id);
|
||||
if (!key) return res.status(404).json({ error: '密钥不存在' });
|
||||
res.json({ ok: true, value: key.value, label: key.label });
|
||||
});
|
||||
|
||||
// 17. 删除密钥
|
||||
app.delete('/api/keys/:id', (req, res) => {
|
||||
keyStore = keyStore.filter(k => k.id !== req.params.id);
|
||||
saveKeys();
|
||||
res.json({ ok: true });
|
||||
});
|
||||
|
||||
// ========== GPU & 训练 & Agent API (v3.1 新增) ==========
|
||||
|
||||
// 18. GPU状态 - Agent推送
|
||||
app.post('/api/gpu/status', (req, res) => {
|
||||
if (!checkAgentAuth(req)) return res.status(401).json({ error: '未授权，请检查API Key' });
|
||||
const data = {
|
||||
timestamp: new Date().toISOString(),
|
||||
hostname: req.body.hostname || '3090-server',
|
||||
gpus: req.body.gpus || []
|
||||
};
|
||||
try {
|
||||
fs.writeFileSync(DATA_DIR + '/gpu-status.json', JSON.stringify(data, null, 2));
|
||||
res.json({ ok: true });
|
||||
} catch(e) {
|
||||
res.status(500).json({ error: '存储失败: ' + e.message });
|
||||
}
|
||||
});
|
||||
|
||||
// 19. GPU状态 - 仪表盘读取
|
||||
app.get('/api/gpu/status', (req, res) => {
|
||||
try {
|
||||
const data = JSON.parse(fs.readFileSync(DATA_DIR + '/gpu-status.json', 'utf-8'));
|
||||
const age = Date.now() - new Date(data.timestamp).getTime();
|
||||
res.json({ ok: true, ...data, online: age < 120000, age_seconds: Math.floor(age / 1000) });
|
||||
} catch(e) {
|
||||
res.json({ ok: true, gpus: [], timestamp: null, online: false });
|
||||
}
|
||||
});
|
||||
|
||||
// 20. 训练状态 - Agent推送
|
||||
app.post('/api/training/status', (req, res) => {
|
||||
if (!checkAgentAuth(req)) return res.status(401).json({ error: '未授权' });
|
||||
const data = {
|
||||
timestamp: new Date().toISOString(),
|
||||
job_id: req.body.job_id || 'hldp-3b-test',
|
||||
status: req.body.status || 'idle',
|
||||
step: req.body.step || 0,
|
||||
total_steps: req.body.total_steps || 0,
|
||||
loss: req.body.loss || null,
|
||||
loss_history: req.body.loss_history || [],
|
||||
eta_seconds: req.body.eta_seconds || null,
|
||||
learning_rate: req.body.learning_rate || null,
|
||||
elapsed_seconds: req.body.elapsed_seconds || 0,
|
||||
model_name: req.body.model_name || '',
|
||||
message: req.body.message || ''
|
||||
};
|
||||
try {
|
||||
fs.writeFileSync(DATA_DIR + '/training-status.json', JSON.stringify(data, null, 2));
|
||||
res.json({ ok: true });
|
||||
} catch(e) {
|
||||
res.status(500).json({ error: '存储失败' });
|
||||
}
|
||||
});
|
||||
|
||||
// 21. 训练状态 - 仪表盘读取
|
||||
app.get('/api/training/status', (req, res) => {
|
||||
try {
|
||||
const data = JSON.parse(fs.readFileSync(DATA_DIR + '/training-status.json', 'utf-8'));
|
||||
const age = Date.now() - new Date(data.timestamp).getTime();
|
||||
res.json({ ok: true, ...data, online: age < 120000 });
|
||||
} catch(e) {
|
||||
res.json({ ok: true, status: 'offline', message: '铸渊Agent未连接' });
|
||||
}
|
||||
});
|
||||
|
||||
// 22. Agent操作日志 - Agent推送
|
||||
app.post('/api/agent/log', (req, res) => {
|
||||
if (!checkAgentAuth(req)) return res.status(401).json({ error: '未授权' });
|
||||
const entry = {
|
||||
timestamp: new Date().toISOString(),
|
||||
level: req.body.level || 'info',
|
||||
category: req.body.category || 'agent',
|
||||
message: req.body.message || ''
|
||||
};
|
||||
try {
|
||||
const logPath = DATA_DIR + '/agent-log.jsonl';
|
||||
fs.appendFileSync(logPath, JSON.stringify(entry) + '\n');
|
||||
const lines = fs.readFileSync(logPath, 'utf-8').trim().split('\n').filter(Boolean);
|
||||
if (lines.length > 500) {
|
||||
fs.writeFileSync(logPath, lines.slice(-500).join('\n') + '\n');
|
||||
}
|
||||
res.json({ ok: true });
|
||||
} catch(e) {
|
||||
res.status(500).json({ error: '存储失败' });
|
||||
}
|
||||
});
|
||||
|
||||
// 23. Agent操作日志 - 仪表盘读取
|
||||
app.get('/api/agent/log', (req, res) => {
|
||||
try {
|
||||
const logPath = DATA_DIR + '/agent-log.jsonl';
|
||||
if (!fs.existsSync(logPath)) return res.json({ ok: true, entries: [], total: 0 });
|
||||
const lines = fs.readFileSync(logPath, 'utf-8').trim().split('\n').filter(Boolean);
|
||||
const limit = parseInt(req.query.limit) || 50;
|
||||
const entries = lines.slice(-limit).map(l => {
|
||||
try { return JSON.parse(l); } catch(e) { return { level: 'info', message: l }; }
|
||||
});
|
||||
res.json({ ok: true, entries, total: lines.length });
|
||||
} catch(e) {
|
||||
res.json({ ok: true, entries: [], total: 0 });
|
||||
}
|
||||
});
|
||||
|
||||
// 24. Agent日记 - Agent推送
|
||||
app.post('/api/agent/diary', (req, res) => {
|
||||
if (!checkAgentAuth(req)) return res.status(401).json({ error: '未授权' });
|
||||
const entry = {
|
||||
timestamp: new Date().toISOString(),
|
||||
type: req.body.type || 'info',
|
||||
title: req.body.title || '',
|
||||
description: req.body.description || ''
|
||||
};
|
||||
try {
|
||||
const diaryPath = DATA_DIR + '/agent-diary.jsonl';
|
||||
fs.appendFileSync(diaryPath, JSON.stringify(entry) + '\n');
|
||||
const lines = fs.readFileSync(diaryPath, 'utf-8').trim().split('\n').filter(Boolean);
|
||||
if (lines.length > 200) {
|
||||
fs.writeFileSync(diaryPath, lines.slice(-200).join('\n') + '\n');
|
||||
}
|
||||
res.json({ ok: true });
|
||||
} catch(e) {
|
||||
res.status(500).json({ error: '存储失败' });
|
||||
}
|
||||
});
|
||||
|
||||
// 25. Agent日记 - 仪表盘读取
|
||||
app.get('/api/agent/diary', (req, res) => {
|
||||
try {
|
||||
const diaryPath = DATA_DIR + '/agent-diary.jsonl';
|
||||
if (!fs.existsSync(diaryPath)) return res.json({ ok: true, entries: [], total: 0 });
|
||||
const lines = fs.readFileSync(diaryPath, 'utf-8').trim().split('\n').filter(Boolean);
|
||||
const entries = lines.map(l => {
|
||||
try { return JSON.parse(l); } catch(e) { return { type: 'info', title: l }; }
|
||||
});
|
||||
res.json({ ok: true, entries: entries.reverse().slice(0, 20), total: entries.length });
|
||||
} catch(e) {
|
||||
res.json({ ok: true, entries: [], total: 0 });
|
||||
}
|
||||
});
|
||||
|
||||
// 26. 综合状态 - 一次性拉取所有Agent数据
|
||||
app.get('/api/agent/combined', (req, res) => {
|
||||
const result = { gpu: null, training: null, logs: [], diary: [], online: false, last_seen: null };
|
||||
try {
|
||||
if (fs.existsSync(DATA_DIR + '/gpu-status.json')) {
|
||||
result.gpu = JSON.parse(fs.readFileSync(DATA_DIR + '/gpu-status.json', 'utf-8'));
|
||||
result.last_seen = result.gpu.timestamp;
|
||||
result.online = result.gpu.timestamp && (Date.now() - new Date(result.gpu.timestamp).getTime() < 120000);
|
||||
}
|
||||
} catch(e) {}
|
||||
try {
|
||||
if (fs.existsSync(DATA_DIR + '/training-status.json')) {
|
||||
result.training = JSON.parse(fs.readFileSync(DATA_DIR + '/training-status.json', 'utf-8'));
|
||||
}
|
||||
} catch(e) {}
|
||||
try {
|
||||
const logPath = DATA_DIR + '/agent-log.jsonl';
|
||||
if (fs.existsSync(logPath)) {
|
||||
const logLines = fs.readFileSync(logPath, 'utf-8').trim().split('\n').filter(Boolean);
|
||||
result.logs = logLines.slice(-20).map(l => { try { return JSON.parse(l); } catch(e) { return { message: l }; } });
|
||||
}
|
||||
} catch(e) {}
|
||||
try {
|
||||
const diaryPath = DATA_DIR + '/agent-diary.jsonl';
|
||||
if (fs.existsSync(diaryPath)) {
|
||||
const diaryLines = fs.readFileSync(diaryPath, 'utf-8').trim().split('\n').filter(Boolean);
|
||||
result.diary = diaryLines.map(l => { try { return JSON.parse(l); } catch(e) { return { type: 'info', title: l }; } }).reverse().slice(0, 10);
|
||||
}
|
||||
} catch(e) {}
|
||||
res.json(result);
|
||||
});
|
||||
|
||||
// ========== 算力池 API ==========
|
||||
|
||||
let cachedPoolStatus = {};
|
||||
|
||||
// 27. 接收gatekeeper上报的节点状态
|
||||
app.post('/api/pool/report', (req, res) => {
|
||||
const { node_code, node_name, cpu_load, mem_total_mb, mem_avail_mb, disk_used_pct, uptime_hours, tasks, spec } = req.body;
|
||||
if (!node_code) return res.status(400).json({ error: '缺少 node_code' });
|
||||
cachedPoolStatus[node_code] = {
|
||||
name: node_name || node_code,
|
||||
online: true,
|
||||
cpu_load: cpu_load || 0,
|
||||
mem_total_mb: mem_total_mb || 0,
|
||||
mem_avail_mb: mem_avail_mb || 0,
|
||||
disk_used_pct: disk_used_pct || 0,
|
||||
uptime_hours: uptime_hours || 0,
|
||||
tasks: tasks || 0,
|
||||
spec: spec || { cpu: 2, mem_gb: 4 },
|
||||
last_seen: new Date().toISOString()
|
||||
};
|
||||
res.json({ ok: true });
|
||||
});
|
||||
|
||||
// 28. 获取算力池汇总状态
|
||||
app.get('/api/pool/status', (req, res) => {
|
||||
const nodes = {};
|
||||
let totalCpu = 0, usedCpu = 0, totalMem = 0, usedMem = 0, onlineNodes = 0;
|
||||
let totalSpec = { cpu: 0, mem_gb: 0 };
|
||||
|
||||
for (const srv of SERVERS) {
|
||||
const cached = cachedPoolStatus[srv.code];
|
||||
if (cached) {
|
||||
nodes[srv.code] = cached;
|
||||
const spec = cached.spec || { cpu: 2, mem_gb: 4 };
|
||||
totalCpu += spec.cpu;
|
||||
usedCpu += cached.cpu_load;
|
||||
totalMem += spec.mem_gb;
|
||||
const memAvailGb = cached.mem_avail_mb / 1024;
|
||||
usedMem += Math.max(0, spec.mem_gb - memAvailGb);
|
||||
onlineNodes++;
|
||||
totalSpec.cpu += spec.cpu;
|
||||
totalSpec.mem_gb += spec.mem_gb;
|
||||
} else {
|
||||
nodes[srv.code] = {
|
||||
name: srv.name,
|
||||
online: false,
|
||||
cpu_load: 0, mem_total_mb: 0, mem_avail_mb: 0,
|
||||
disk_used_pct: 0, uptime_hours: 0, tasks: 0,
|
||||
spec: { cpu: 2, mem_gb: 4 },
|
||||
last_seen: null
|
||||
};
|
||||
totalSpec.cpu += 2;
|
||||
totalSpec.mem_gb += 4;
|
||||
}
|
||||
}
|
||||
|
||||
const pool = {
|
||||
updated_at: new Date().toISOString(),
|
||||
cpu_total: totalSpec.cpu,
|
||||
cpu_used: usedCpu,
|
||||
cpu_available: Math.max(0, totalSpec.cpu - usedCpu),
|
||||
mem_total_gb: totalSpec.mem_gb,
|
||||
mem_used_gb: usedMem,
|
||||
mem_available_gb: Math.max(0, totalSpec.mem_gb - usedMem),
|
||||
total_nodes: SERVERS.length,
|
||||
online_nodes: onlineNodes
|
||||
};
|
||||
|
||||
res.json({
|
||||
ok: true,
|
||||
pool: { pool, nodes, total_spec: totalSpec, active_tasks: 0, updated_at: pool.updated_at }
|
||||
});
|
||||
});
|
||||
|
||||
const PORT = process.env.CONSOLE_PORT || 3920;
|
||||
app.listen(PORT, '127.0.0.1', () => {
|
||||
console.log(`光湖服务器主控台 API v3.0 — 操作回执系统`);
|
||||
console.log(` 监听: 127.0.0.1:${PORT}`);
|
||||
console.log(` 服务器: ${SERVERS.length} 台`);
|
||||
console.log(` MCP终端: 已启用`);
|
||||
console.log('光湖主控台 API v3.1 — GPU+Agent训练台');
|
||||
console.log(' 监听: 127.0.0.1:' + PORT);
|
||||
console.log(' 服务器: ' + SERVERS.length + ' 台 | GPU训练台: 已启用');
|
||||
console.log(' 数据目录: ' + DATA_DIR);
|
||||
});
|
||||
Loading…
x
Reference in New Issue
Block a user