feat: ModelRouter — unified OpenAI-compatible multi-source LLM gateway

- Lua adapters per upstream (transform_request/response/stream_chunk, build_headers signing hooks)
- AUTO priority routing with per-model kind (chat/image), explicit source/model routing
- Per-source concurrency caps with queueing, exponential backoff, AUTO failover
- OpenAI-compatible API: chat completions, SSE streaming, image generations, models
- Gateway key auth, web UI for adapter/source management, runtime persistence
- e2e test running the real binary against mocked upstreams
This commit is contained in:
root
2026-08-05 15:24:51 +08:00
parent 8631b08253
commit f7f76e097d
32 changed files with 4382 additions and 2 deletions

View File

@ -0,0 +1,241 @@
<!DOCTYPE html>
<html lang="zh">
<head>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1">
<title>llmsproxy</title>
<style>
:root { --bg:#0f1115; --card:#171a21; --line:#262b36; --fg:#e6e8ee; --muted:#8b93a5;
--accent:#4f7cff; --ok:#3ecf8e; --warn:#ffb454; --err:#ff5d6c; }
* { box-sizing:border-box; }
body { margin:0; font:14px/1.5 ui-monospace,Menlo,Consolas,monospace; background:var(--bg); color:var(--fg); }
header { display:flex; align-items:center; gap:12px; padding:14px 20px; border-bottom:1px solid var(--line); }
header h1 { font-size:16px; margin:0; }
header .badge { font-size:12px; color:var(--muted); }
nav { display:flex; gap:4px; padding:10px 20px; border-bottom:1px solid var(--line); }
nav button { background:transparent; border:1px solid transparent; color:var(--muted); padding:6px 14px;
cursor:pointer; border-radius:6px; font:inherit; }
nav button.active { background:var(--card); border-color:var(--line); color:var(--fg); }
main { padding:20px; max-width:1200px; margin:0 auto; }
.card { background:var(--card); border:1px solid var(--line); border-radius:10px; padding:16px; margin-bottom:16px; }
.card h2 { font-size:14px; margin:0 0 12px; color:var(--muted); font-weight:600; }
table { width:100%; border-collapse:collapse; }
th,td { text-align:left; padding:8px 10px; border-bottom:1px solid var(--line); font-size:13px; }
th { color:var(--muted); font-weight:500; }
.tag { display:inline-block; padding:2px 8px; border-radius:10px; font-size:11px; margin:2px; }
.tag-green { background:#143b2b; color:var(--ok); }
.tag-red { background:#3b1418; color:var(--err); }
.tag-blue { background:#14223b; color:var(--accent); }
button { background:var(--accent); color:#fff; border:0; border-radius:6px; padding:7px 14px; cursor:pointer; font:inherit; }
button.ghost { background:transparent; border:1px solid var(--line); color:var(--muted); }
button.danger { background:transparent; border:1px solid #3b1418; color:var(--err); }
input,select,textarea { width:100%; background:#10131a; border:1px solid var(--line); color:var(--fg);
border-radius:6px; padding:7px 10px; font:inherit; margin-bottom:8px; }
textarea { min-height:220px; resize:vertical; }
label { display:block; font-size:12px; color:var(--muted); margin:10px 0 4px; }
.row { display:flex; gap:12px; } .row > div { flex:1; }
.model-row { display:flex; gap:8px; align-items:center; }
.model-row input { margin:0; } .model-row .del { flex:0 0 auto; padding:4px 8px; }
.muted { color:var(--muted); }
.hidden { display:none; }
#toast { position:fixed; bottom:20px; right:20px; background:var(--card); border:1px solid var(--line);
padding:10px 16px; border-radius:8px; display:none; }
</style>
</head>
<body>
<header>
<h1>llmsproxy</h1>
<span class="badge">统一 LLM 网关 · 适配器/源管理</span>
</header>
<nav>
<button data-tab="status" class="active">状态</button>
<button data-tab="sources"></button>
<button data-tab="adapters">适配器</button>
</nav>
<main>
<div id="tab-status"></div>
<div id="tab-sources" class="hidden"></div>
<div id="tab-adapters" class="hidden"></div>
</main>
<div id="toast"></div>
<script>
const $ = s => document.querySelector(s);
const tab = () => document.querySelector('nav button.active').dataset.tab;
const api = (p, o) => fetch(p, o).then(async r => {
const j = await r.json().catch(() => ({}));
if (!r.ok) throw new Error((j.error && j.error.message) || r.statusText);
return j;
});
function toast(m) { const t = $('#toast'); t.textContent = m; t.style.display = 'block'; setTimeout(() => t.style.display = 'none', 3000); }
document.querySelectorAll('nav button').forEach(b => b.onclick = () => {
document.querySelectorAll('nav button').forEach(x => x.classList.toggle('active', x === b));
['status','sources','adapters'].forEach(t => $('#tab-' + t).classList.toggle('hidden', t !== b.dataset.tab));
refresh(b.dataset.tab);
});
async function refresh(t) {
if (t === 'status') return renderStatus();
if (t === 'sources') return renderSources();
return renderAdapters();
}
async function renderStatus() {
const s = await api('/api/status');
const src = s.sources.map(x => `<tr><td>${esc(x.name)}</td><td>${esc(x.adapter)}</td>
<td>${x.models.map(m => `<span class="tag tag-blue">${esc(m)}</span>`).join('')}</td>
<td>${x.available ? '<span class="tag tag-green">可用</span>' : '<span class="tag tag-red">退避/不可用</span>'}</td>
<td>${x.max_concurrent}</td></tr>`).join('');
$('#tab-status').innerHTML = `
<div class="card"><h2>网关</h2>
<div class="muted">默认模型: ${esc(s.default_model)}</div>
<div class="muted">模型列表: ${s.models.map(esc).join(', ')}</div>
</div>
<div class="card"><h2>源状态 (${s.sources.length})</h2>
<table><tr><th>名称</th><th>适配器</th><th>模型</th><th>健康</th><th>并发</th></tr>${src}</table>
</div>
<div class="card"><h2>已加载适配器 (${s.adapters.length})</h2>
<table><tr><th>名称</th><th>版本</th></tr>
${s.adapters.map(a => `<tr><td>${esc(a.name)}</td><td>${esc(a.version || '')}</td></tr>`).join('')}
</table></div>`;
}
async function renderSources() {
const j = await api('/api/sources');
const rows = j.sources.map(s => `<tr><td>${esc(s.name)}</td><td>${esc(s.base_url)}</td><td>${esc(s.adapter)}</td>
<td>${s.models.map(m => `<span class="tag tag-blue">${esc(m.id)}<span class="muted">·${m.priority||0}</span></span>`).join('')}</td>
<td><button class="ghost" onclick="editSource(${JSON.stringify(s.name).replace(/"/g,'&quot;')})">编辑</button>
<button class="danger" onclick="delSource('${escAttr(s.name)}')">删除</button></td></tr>`).join('');
$('#tab-sources').innerHTML = `
<div class="card"><h2>源 (${j.sources.length})</h2>
<table><tr><th>名称</th><th>地址</th><th>适配器</th><th>模型 (优先级)</th><th></th></tr>${rows}</table>
<p><button onclick="editSource('')">+ 新增源</button></p>
</div>`;
}
async function renderAdapters() {
const j = await api('/api/status');
const rows = j.adapters.map(a => `<tr><td>${esc(a.name)}</td><td>${esc(a.version || '')}</td>
<td><button class="danger" onclick="delAdapter('${escAttr(a.name)}')">删除</button></td></tr>`).join('');
$('#tab-adapters').innerHTML = `
<div class="card"><h2>已加载适配器 (${j.adapters.length})</h2>
<table><tr><th>名称</th><th>版本</th><th></th></tr>${rows}</table>
</div>
<div class="card"><h2>上传 Lua 适配器</h2>
<label>名称(脚本保存为 <code>&lt;name&gt;.lua</code></label>
<input id="adp-name" placeholder="如 mysrc">
<label>Lua 脚本(返回 adapter table支持 transform_request/response/stream_chunk/build_headers</label>
<textarea id="adp-code" spellcheck="false" placeholder="return { name='mysrc', endpoint='/chat/completions', transform_request=function(raw) return raw end, transform_response=function(raw) return raw end }"></textarea>
<p><button onclick="uploadAdapter()">上传并加载</button></p>
</div>`;
}
function editSource(name) {
const modal = document.createElement('div');
const existing = name ? null : null;
// We'll re-fetch and prefill
api('/api/sources').then(j => {
const s = j.sources.find(x => x.name === name) || { name: name, models: [{ id: '', priority: 0, kind: 'chat' }] };
const modelInputs = (s.models || []).map((m, i) => modelRow(m, i)).join('');
modal.innerHTML = `<div class="card"><h2>${name ? '编辑源: ' + esc(name) : '新增源'}</h2>
<label>名称</label><input id="s-name" value="${escAttr(s.name)}" ${name ? 'disabled' : ''}>
<label>Base URL</label><input id="s-url" value="${escAttr(s.base_url || '')}">
<label>API Key</label><input id="s-key" type="password" value="${escAttr(s.api_key || '')}">
<label>适配器(对应已加载的 Lua 适配器名)</label><input id="s-adapter" value="${escAttr(s.adapter || 'openai')}">
<div class="row">
<div><label>聊天端点 (可选覆盖)</label><input id="s-ep" value="${escAttr(s.endpoint || '')}"></div>
<div><label>生图端点 (可选覆盖)</label><input id="s-img" value="${escAttr(s.image_endpoint || '')}"></div>
</div>
<div class="row">
<div><label>并发上限</label><input id="s-conc" type="number" value="${s.max_concurrent || 8}"></div>
<div><label>温度</label><input id="s-temp" type="number" step="0.1" value="${s.temperature || 0.7}"></div>
</div>
<label>模型列表(优先级数字越大越优先被 AUTO 选中)</label>
<div id="s-models">${modelInputs}</div>
<button class="ghost" onclick="addModelRow()">+ 模型</button>
<label>Meta透传给 build_headers 钩子JSON</label>
<textarea id="s-meta" style="min-height:80px" placeholder='{"app_id":"x","app_secret":"y"}'>${esc(JSON.stringify(s.meta || {}, null, 2))}</textarea>
<p><button onclick="saveSource(this)">保存</button> <button class="ghost" onclick="modal.remove()">取消</button></p>
</div>`;
modal.style.cssText = 'position:fixed;inset:0;background:rgba(0,0,0,.6);display:flex;align-items:flex-start;justify-content:center;overflow:auto;padding:40px 20px;z-index:50';
modal.id = 'modal';
document.body.appendChild(modal);
window._modal = modal;
window._models = s.models || [];
});
}
function modelRow(m, i) {
return `<div class="model-row">
<input data-mi="${i}" class="m-id" placeholder="模型 id如 deepseek-v4-flash" value="${escAttr(m.id)}">
<input data-mi="${i}" class="m-prio" type="number" placeholder="优先级" value="${m.priority || 0}" style="width:90px">
<select data-mi="${i}" class="m-kind"><option ${(m.kind==='image')?'':'selected'} value="chat">chat</option><option ${(m.kind==='image')?'selected':''} value="image">image</option></select>
<button class="ghost del" data-mi="${i}" onclick="this.closest('.model-row').remove()">×</button>
</div>`;
}
function addModelRow() {
const div = $('#s-models');
div.insertAdjacentHTML('beforeend', modelRow({ id: '', priority: 0, kind: 'chat' }, div.children.length));
}
async function saveSource(btn) {
const models = [...document.querySelectorAll('#s-models .model-row')].map(row => ({
id: row.querySelector('.m-id').value.trim(),
priority: parseInt(row.querySelector('.m-prio').value) || 0,
kind: row.querySelector('.m-kind').value,
})).filter(m => m.id);
let meta = {};
try { meta = JSON.parse($('#s-meta').value || '{}'); } catch (e) { toast('Meta 不是合法 JSON'); return; }
const payload = {
name: $('#s-name').value.trim(),
base_url: $('#s-url').value.trim(),
api_key: $('#s-key').value.trim(),
adapter: $('#s-adapter').value.trim(),
endpoint: $('#s-ep').value.trim(),
image_endpoint: $('#s-img').value.trim(),
max_concurrent: parseInt($('#s-conc').value) || 8,
temperature: parseFloat($('#s-temp').value) || 0,
models, meta,
};
btn.disabled = true;
try {
await api('/api/sources', { method: 'POST', body: JSON.stringify(payload) });
toast('已保存并热重载');
window._modal && window._modal.remove();
renderSources();
} catch (e) { toast('保存失败: ' + e.message); btn.disabled = false; }
}
async function delSource(name) {
if (!confirm('删除源 ' + name + '?')) return;
await api('/api/sources/' + encodeURIComponent(name), { method: 'DELETE' });
toast('已删除');
renderSources();
}
async function uploadAdapter() {
const name = $('#adp-name').value.trim();
const code = $('#adp-code').value;
if (!name || !code) return toast('需要名称和脚本');
try {
await api('/api/adapters', { method: 'POST', body: JSON.stringify({ name, code }) });
toast('适配器已加载');
renderAdapters();
} catch (e) { toast('上传失败: ' + e.message); }
}
async function delAdapter(name) {
if (!confirm('删除适配器 ' + name + '?')) return;
await api('/api/adapters/' + encodeURIComponent(name), { method: 'DELETE' });
toast('已删除');
renderAdapters();
}
function esc(s) { return String(s ?? '').replace(/[&<>"']/g, c => ({'&':'&amp;','<':'&lt;','>':'&gt;','"':'&quot;',"'":'&#39;'}[c])); }
function escAttr(s) { return esc(s).replace(/"/g, '&quot;'); }
refresh('status');
</script>
</body>
</html>