feat: add external agent run worker entry

This commit is contained in:
John
2026-07-02 08:02:03 +08:00
parent 3ea973968e
commit 05d29ce62d
9 changed files with 7190 additions and 5 deletions
+5
View File
@@ -29,6 +29,11 @@ H5_PUBLIC_BASE_URL=http://127.0.0.1:5173
# Tool Gateway Queue v0/agent/runs 后台任务队列,默认保守限流)。
# MEMIND_AGENT_RUN_QUEUE_CONCURRENCY=1
# MEMIND_AGENT_RUN_TIMEOUT_MS=900000
# P6.1 外部 Tool Worker 接管入口。默认 Portal 内部 auto-dispatch 保持开启;
# 只有外部 worker 验证就绪后,才把 Portal 设为 0 并独立运行 scripts/agent-run-worker.mjs。
# MEMIND_AGENT_RUN_AUTODISPATCH=1
# MEMIND_AGENT_RUN_WORKER_POLL_MS=1000
# MEMIND_AGENT_RUN_WORKER_BATCH_SIZE=1
# ===== memind_adm(独立后台服务)=====
# 平台超管 /admin-api/* 与广场运营 /api/ops/v1/* 已从公网服务拆出,独立进程运行。
+2
View File
@@ -47,3 +47,5 @@ Streaming runtime operations:
node scripts/check-tool-runtime.mjs
node scripts/check-agent-code-run-entry.mjs
curl -sk https://mm.tkmind.cn/api/runtime/status # includes toolRuntime.queue
node scripts/agent-run-worker.mjs --status
node scripts/agent-run-worker.mjs --once
File diff suppressed because it is too large Load Diff
+42 -4
View File
@@ -6534,6 +6534,11 @@ function timeoutError(ms) {
err.code = "AGENT_RUN_TIMEOUT";
return err;
}
function envFlag(value, fallback = false) {
const raw = String(value ?? "").trim().toLowerCase();
if (!raw) return fallback;
return ["1", "true", "yes", "on"].includes(raw);
}
function normalizeAgentRunToolMode(value) {
const normalized = String(value ?? "chat").trim().toLowerCase();
if (!normalized || normalized === "chat") return "chat";
@@ -6589,7 +6594,7 @@ function createAgentRunGateway({
userAuth: userAuth2,
tkmindProxy: tkmindProxy2,
retryDelaysMs = DEFAULT_RUN_RETRY_DELAYS_MS,
autoDispatch = true,
autoDispatch = envFlag(process.env.MEMIND_AGENT_RUN_AUTODISPATCH, true),
maxConcurrentRuns = positiveInteger(
process.env.MEMIND_AGENT_RUN_QUEUE_CONCURRENCY,
DEFAULT_MAX_CONCURRENT_RUNS
@@ -6810,6 +6815,7 @@ function createAgentRunGateway({
statusCounts[row.status] = Number(row.count ?? 0);
}
return {
autoDispatch,
maxConcurrentRuns,
runTimeoutMs,
inFlight: inFlight.size,
@@ -6818,22 +6824,51 @@ function createAgentRunGateway({
terminalStatuses: [...TERMINAL_STATUSES]
};
}
async function dispatchQueuedRuns({ limit = maxConcurrentRuns } = {}) {
const dispatchLimit = Math.max(1, Number(limit) || maxConcurrentRuns);
const available = Math.max(0, maxConcurrentRuns - inFlight.size - queuedDispatches.length);
if (available <= 0) {
return {
considered: 0,
dispatched: 0,
queue: await getQueueStatus()
};
}
const take = Math.min(dispatchLimit, available);
const [rows] = await pool2.query(
`SELECT id
FROM h5_agent_runs
WHERE status IN ('queued', 'retryable')
ORDER BY updated_at ASC
LIMIT ?`,
[take]
);
for (const row of rows) {
dispatchRun(row.id);
}
return {
considered: rows.length,
dispatched: rows.length,
queue: await getQueueStatus()
};
}
return {
createRun,
getRunForUser,
dispatchRun,
getQueueStatus
getQueueStatus,
dispatchQueuedRuns
};
}
// agent-run-routes.mjs
function envFlag(value) {
function envFlag2(value) {
return ["1", "true", "yes", "on"].includes(String(value ?? "").trim().toLowerCase());
}
function createPostAgentRunsHandler({
userAuth: userAuth2,
agentRunGateway: agentRunGateway2,
codeRunsEnabled = envFlag(process.env.MEMIND_AGENT_CODE_RUNS_ENABLED)
codeRunsEnabled = envFlag2(process.env.MEMIND_AGENT_CODE_RUNS_ENABLED)
}) {
return async function postAgentRuns(request, response) {
try {
@@ -35041,6 +35076,9 @@ async function bootstrapUserAuth() {
pool: pool2,
userAuth,
tkmindProxy,
autoDispatch: ["1", "true", "yes", "on"].includes(
String(process.env.MEMIND_AGENT_RUN_AUTODISPATCH ?? "1").trim().toLowerCase()
),
maxConcurrentRuns: Number(process.env.MEMIND_AGENT_RUN_QUEUE_CONCURRENCY ?? 1),
runTimeoutMs: Number(process.env.MEMIND_AGENT_RUN_TIMEOUT_MS ?? 15 * 60 * 1e3)
});
+38 -1
View File
@@ -35,6 +35,12 @@ function timeoutError(ms) {
return err;
}
function envFlag(value, fallback = false) {
const raw = String(value ?? '').trim().toLowerCase();
if (!raw) return fallback;
return ['1', 'true', 'yes', 'on'].includes(raw);
}
export function normalizeAgentRunToolMode(value) {
const normalized = String(value ?? 'chat').trim().toLowerCase();
if (!normalized || normalized === 'chat') return 'chat';
@@ -101,7 +107,7 @@ export function createAgentRunGateway({
userAuth,
tkmindProxy,
retryDelaysMs = DEFAULT_RUN_RETRY_DELAYS_MS,
autoDispatch = true,
autoDispatch = envFlag(process.env.MEMIND_AGENT_RUN_AUTODISPATCH, true),
maxConcurrentRuns = positiveInteger(
process.env.MEMIND_AGENT_RUN_QUEUE_CONCURRENCY,
DEFAULT_MAX_CONCURRENT_RUNS,
@@ -340,6 +346,7 @@ export function createAgentRunGateway({
statusCounts[row.status] = Number(row.count ?? 0);
}
return {
autoDispatch,
maxConcurrentRuns,
runTimeoutMs,
inFlight: inFlight.size,
@@ -349,10 +356,40 @@ export function createAgentRunGateway({
};
}
async function dispatchQueuedRuns({ limit = maxConcurrentRuns } = {}) {
const dispatchLimit = Math.max(1, Number(limit) || maxConcurrentRuns);
const available = Math.max(0, maxConcurrentRuns - inFlight.size - queuedDispatches.length);
if (available <= 0) {
return {
considered: 0,
dispatched: 0,
queue: await getQueueStatus(),
};
}
const take = Math.min(dispatchLimit, available);
const [rows] = await pool.query(
`SELECT id
FROM h5_agent_runs
WHERE status IN ('queued', 'retryable')
ORDER BY updated_at ASC
LIMIT ?`,
[take],
);
for (const row of rows) {
dispatchRun(row.id);
}
return {
considered: rows.length,
dispatched: rows.length,
queue: await getQueueStatus(),
};
}
return {
createRun,
getRunForUser,
dispatchRun,
getQueueStatus,
dispatchQueuedRuns,
};
}
+82
View File
@@ -30,6 +30,14 @@ function createFakePool() {
}
return [[...counts].map(([status, count]) => ({ status, count }))];
}
if (sql.includes('SELECT id') && sql.includes("status IN ('queued', 'retryable')")) {
const limit = Number(params[0] ?? 1);
return [[...runs.values()]
.filter((row) => ['queued', 'retryable'].includes(row.status))
.sort((a, b) => Number(a.updated_at) - Number(b.updated_at))
.slice(0, limit)
.map((row) => ({ id: row.id }))];
}
if (sql.includes('INSERT INTO h5_agent_runs')) {
const [
id,
@@ -338,3 +346,77 @@ test('agent run queue status reports active database and local queue state', asy
assert.equal(status.pendingDispatches, 0);
assert.equal(status.statusCounts.queued, 1);
});
test('external worker dispatches queued runs through the same queue controls', async () => {
const pool = createFakePool();
const submitted = [];
const gateway = createAgentRunGateway({
pool,
userAuth: {},
tkmindProxy: {
async startSessionForUser() {
return { id: 'session-worker' };
},
async submitSessionReplyForUser(_userId, _sessionId, requestId) {
submitted.push(requestId);
},
},
autoDispatch: false,
retryDelaysMs: [],
maxConcurrentRuns: 1,
});
const run = await gateway.createRun('user-1', {
requestId: 'req-worker',
userMessage: { role: 'user', content: [] },
});
assert.equal(pool.runs.get(run.id).status, 'queued');
const result = await gateway.dispatchQueuedRuns({ limit: 10 });
assert.equal(result.dispatched, 1);
await waitFor(() => pool.runs.get(run.id)?.status === 'succeeded');
assert.deepEqual(submitted, ['req-worker']);
});
test('external worker does not dispatch more runs when local queue is full', async () => {
const pool = createFakePool();
const release = [];
const gateway = createAgentRunGateway({
pool,
userAuth: {},
tkmindProxy: {
async startSessionForUser() {
return { id: `session-${release.length + 1}` };
},
async submitSessionReplyForUser() {
await new Promise((resolve) => release.push(resolve));
},
},
autoDispatch: false,
retryDelaysMs: [],
maxConcurrentRuns: 1,
});
const run1 = await gateway.createRun('user-1', {
requestId: 'req-full-1',
userMessage: { role: 'user', content: [] },
});
const run2 = await gateway.createRun('user-1', {
requestId: 'req-full-2',
userMessage: { role: 'user', content: [] },
});
try {
const first = await gateway.dispatchQueuedRuns({ limit: 10 });
assert.equal(first.dispatched, 1);
await waitFor(() => pool.runs.get(run1.id)?.status === 'running');
assert.equal((await gateway.getQueueStatus()).inFlight, 1);
const second = await gateway.dispatchQueuedRuns({ limit: 10 });
assert.equal(second.dispatched, 0);
assert.equal(pool.runs.get(run2.id).status, 'queued');
} finally {
while (release.length > 0) release.shift()();
}
await waitFor(() => pool.runs.get(run1.id)?.status === 'succeeded');
assert.equal(pool.runs.get(run2.id).status, 'queued');
});
+121
View File
@@ -0,0 +1,121 @@
#!/usr/bin/env node
import fs from 'node:fs';
import path from 'node:path';
import { fileURLToPath } from 'node:url';
import { createAgentRunGateway } from '../agent-run-gateway.mjs';
import { createDbPool } from '../db.mjs';
import { createTkmindProxy } from '../tkmind-proxy.mjs';
import { createUserAuth } from '../user-auth.mjs';
const root = path.join(path.dirname(fileURLToPath(import.meta.url)), '..');
function loadEnvFile(filePath) {
if (!fs.existsSync(filePath)) return;
for (const line of fs.readFileSync(filePath, 'utf8').split('\n')) {
const trimmed = line.trim();
if (!trimmed || trimmed.startsWith('#')) continue;
const eq = trimmed.indexOf('=');
if (eq < 0) continue;
const key = trimmed.slice(0, eq).trim();
const value = trimmed.slice(eq + 1).trim();
if (!process.env[key]) process.env[key] = value;
}
}
function parseTargets() {
const csv = String(process.env.TKMIND_API_TARGETS ?? '').trim();
if (csv) {
return csv.split(',').map((item) => item.trim()).filter(Boolean);
}
return [process.env.TKMIND_API_TARGET ?? 'https://127.0.0.1:18006'];
}
function parseArgs(argv) {
const args = {
once: false,
status: false,
pollMs: Number(process.env.MEMIND_AGENT_RUN_WORKER_POLL_MS ?? 1000),
limit: Number(process.env.MEMIND_AGENT_RUN_WORKER_BATCH_SIZE ?? 1),
};
for (let i = 0; i < argv.length; i += 1) {
const item = argv[i];
if (item === '--once') args.once = true;
else if (item === '--status') args.status = true;
else if (item === '--poll-ms') args.pollMs = Number(argv[++i] ?? args.pollMs);
else if (item === '--limit') args.limit = Number(argv[++i] ?? args.limit);
else if (item === '--help' || item === '-h') args.help = true;
}
return args;
}
function printHelp() {
console.log([
'Usage:',
' node scripts/agent-run-worker.mjs [--once] [--status] [--poll-ms 1000] [--limit 1]',
'',
'Notes:',
' - Processes existing h5_agent_runs queued/retryable rows through the Tool Gateway queue.',
' - Does not create agent runs by itself.',
' - Set MEMIND_AGENT_RUN_AUTODISPATCH=0 on Portal only when this worker is ready to take over.',
].join('\n'));
}
loadEnvFile(process.env.MEMIND_ENV_FILE || path.join(root, '.env'));
loadEnvFile(path.join(root, '.env.local'));
const args = parseArgs(process.argv.slice(2));
if (args.help) {
printHelp();
process.exit(0);
}
const pool = createDbPool();
const userAuth = createUserAuth(pool, {
usersRoot: process.env.H5_USERS_ROOT ?? path.join(root, 'users'),
h5Root: root,
});
const apiTargets = parseTargets();
const tkmindProxy = createTkmindProxy({
apiTarget: apiTargets[0],
apiTargets,
apiSecret: process.env.TKMIND_SERVER__SECRET_KEY ?? 'local-dev-secret',
userAuth,
});
const gateway = createAgentRunGateway({
pool,
userAuth,
tkmindProxy,
autoDispatch: false,
maxConcurrentRuns: Number(process.env.MEMIND_AGENT_RUN_QUEUE_CONCURRENCY ?? 1),
runTimeoutMs: Number(process.env.MEMIND_AGENT_RUN_TIMEOUT_MS ?? 15 * 60 * 1000),
});
let stopping = false;
process.on('SIGINT', () => { stopping = true; });
process.on('SIGTERM', () => { stopping = true; });
async function runOnce() {
const result = await gateway.dispatchQueuedRuns({ limit: args.limit });
console.log(JSON.stringify({
ok: true,
mode: args.status ? 'status' : 'dispatch',
dispatched: args.status ? 0 : result.dispatched,
queue: result.queue,
}));
}
try {
if (args.status) {
console.log(JSON.stringify({ ok: true, queue: await gateway.getQueueStatus() }, null, 2));
} else if (args.once) {
await runOnce();
} else {
console.log(JSON.stringify({ ok: true, worker: 'agent-run-worker', pollMs: args.pollMs, limit: args.limit }));
while (!stopping) {
await runOnce();
await new Promise((resolve) => setTimeout(resolve, Math.max(200, args.pollMs)));
}
}
} finally {
await pool.end().catch(() => {});
}
+24
View File
@@ -119,6 +119,26 @@ async function bundleSandboxMcp() {
await run(esbuildBin, args);
}
async function bundleAgentRunWorker() {
if (!(await exists(esbuildBin))) {
throw new Error(`未找到 esbuild: ${esbuildBin}`);
}
console.log('==> 打包 agent run worker 为单文件 runtime');
const args = [
'scripts/agent-run-worker.mjs',
'--bundle',
'--platform=node',
'--format=esm',
`--target=${runtimeNodeTarget}`,
'--outfile=.runtime/portal/scripts/agent-run-worker.mjs',
'--banner:js=import { createRequire as __createRequire } from "node:module"; const require = __createRequire(import.meta.url);',
];
for (const pkg of externalPackages) {
args.push(`--external:${pkg}`);
}
await run(esbuildBin, args);
}
async function copyRuntimeAssets() {
console.log('==> 拷贝运行时静态资源');
await remove(runtimeRoot);
@@ -347,6 +367,8 @@ async function writeMetadata() {
' node scripts/check-tool-runtime.mjs',
' node scripts/check-agent-code-run-entry.mjs',
' curl -sk https://mm.tkmind.cn/api/runtime/status # includes toolRuntime.queue',
' node scripts/agent-run-worker.mjs --status',
' node scripts/agent-run-worker.mjs --once',
'',
].join('\n'),
);
@@ -357,6 +379,7 @@ async function main() {
await copyRuntimeAssets();
await bundleServer();
await bundleSandboxMcp();
await bundleAgentRunWorker();
if (runtimeInstallMode === 'bundle-node-modules' && !skipNodeModules) {
await copyNodeModules();
}
@@ -368,6 +391,7 @@ async function main() {
await fs.chmod(path.join(runtimeRoot, 'scripts', 'runtime-worker-drain.mjs'), 0o755);
await fs.chmod(path.join(runtimeRoot, 'scripts', 'runtime-worker-metrics.mjs'), 0o755);
await fs.chmod(path.join(runtimeRoot, 'scripts', 'runtime-slo-report.mjs'), 0o755);
await fs.chmod(path.join(runtimeRoot, 'scripts', 'agent-run-worker.mjs'), 0o755);
await fs.chmod(path.join(runtimeRoot, 'scripts', 'install-runtime-metrics-agent.sh'), 0o755);
await fs.chmod(path.join(runtimeRoot, 'scripts', 'check-tool-runtime.mjs'), 0o755);
await fs.chmod(path.join(runtimeRoot, 'scripts', 'memind-portal-tunnel.sh'), 0o755);
+3
View File
@@ -572,6 +572,9 @@ async function bootstrapUserAuth() {
pool,
userAuth,
tkmindProxy,
autoDispatch: ['1', 'true', 'yes', 'on'].includes(
String(process.env.MEMIND_AGENT_RUN_AUTODISPATCH ?? '1').trim().toLowerCase(),
),
maxConcurrentRuns: Number(process.env.MEMIND_AGENT_RUN_QUEUE_CONCURRENCY ?? 1),
runTimeoutMs: Number(process.env.MEMIND_AGENT_RUN_TIMEOUT_MS ?? 15 * 60 * 1000),
});