feat: harden orchestrator execution runtime
This commit is contained in:
@@ -22,9 +22,49 @@ function authorize(serviceToken) {
|
||||
};
|
||||
}
|
||||
|
||||
function authorizeWorker(workerToken) {
|
||||
const expected = String(workerToken ?? '').trim();
|
||||
return (request, response, next) => {
|
||||
if (!expected) return next();
|
||||
if (request.get('x-orchestrator-worker-token') === expected) return next();
|
||||
return response.status(401).json({
|
||||
error: {
|
||||
code: 'WORKER_UNAUTHORIZED',
|
||||
message: 'Missing or invalid worker token',
|
||||
},
|
||||
});
|
||||
};
|
||||
}
|
||||
|
||||
function prometheusMetrics(metrics) {
|
||||
const lines = [
|
||||
'# HELP memind_orchestrator_executor_jobs Executor jobs by state.',
|
||||
'# TYPE memind_orchestrator_executor_jobs gauge',
|
||||
];
|
||||
for (const [status, count] of Object.entries(metrics.queue?.counts ?? {})) {
|
||||
lines.push(`memind_orchestrator_executor_jobs{status="${status}"} ${Number(count) || 0}`);
|
||||
}
|
||||
lines.push(
|
||||
'# HELP memind_orchestrator_executor_claimable Claimable executor jobs.',
|
||||
'# TYPE memind_orchestrator_executor_claimable gauge',
|
||||
`memind_orchestrator_executor_claimable ${Number(metrics.queue?.claimable) || 0}`,
|
||||
'# HELP memind_orchestrator_executor_expired_leases Expired executor leases.',
|
||||
'# TYPE memind_orchestrator_executor_expired_leases gauge',
|
||||
`memind_orchestrator_executor_expired_leases ${Number(metrics.queue?.expiredLeases) || 0}`,
|
||||
'# HELP memind_orchestrator_executor_workers Healthy registered workers.',
|
||||
'# TYPE memind_orchestrator_executor_workers gauge',
|
||||
`memind_orchestrator_executor_workers ${Number(metrics.queue?.healthyWorkers) || 0}`,
|
||||
'# HELP memind_orchestrator_execution_enabled Whether executor queue admission is enabled.',
|
||||
'# TYPE memind_orchestrator_execution_enabled gauge',
|
||||
`memind_orchestrator_execution_enabled ${metrics.gateway?.executionEnabled ? 1 : 0}`,
|
||||
);
|
||||
return `${lines.join('\n')}\n`;
|
||||
}
|
||||
|
||||
export function createOrchestratorApp({
|
||||
runtime,
|
||||
serviceToken = process.env.MEMIND_ORCHESTRATOR_SERVICE_TOKEN,
|
||||
workerToken = process.env.MEMIND_ORCHESTRATOR_WORKER_TOKEN,
|
||||
} = {}) {
|
||||
if (!runtime) throw new Error('Orchestrator app requires runtime');
|
||||
const app = express();
|
||||
@@ -42,6 +82,16 @@ export function createOrchestratorApp({
|
||||
});
|
||||
}
|
||||
});
|
||||
app.get('/metrics', async (_request, response, next) => {
|
||||
try {
|
||||
const metrics = await runtime.metrics();
|
||||
return response
|
||||
.type('text/plain; version=0.0.4; charset=utf-8')
|
||||
.send(prometheusMetrics(metrics));
|
||||
} catch (error) {
|
||||
return next(error);
|
||||
}
|
||||
});
|
||||
|
||||
app.use('/v1', authorize(serviceToken));
|
||||
|
||||
@@ -53,6 +103,15 @@ export function createOrchestratorApp({
|
||||
}
|
||||
});
|
||||
|
||||
app.post('/v1/executor-jobs', async (request, response, next) => {
|
||||
try {
|
||||
const result = await runtime.submitExecutorJob(request.body);
|
||||
return response.status(result.created ? 202 : 200).json(result);
|
||||
} catch (error) {
|
||||
return next(error);
|
||||
}
|
||||
});
|
||||
|
||||
app.get('/v1/runs/:runId', async (request, response, next) => {
|
||||
try {
|
||||
const state = await runtime.getState(request.params.runId);
|
||||
@@ -122,6 +181,105 @@ export function createOrchestratorApp({
|
||||
}
|
||||
});
|
||||
|
||||
const requireWorker = authorizeWorker(workerToken);
|
||||
|
||||
app.post('/v1/workers/claim', requireWorker, async (request, response, next) => {
|
||||
try {
|
||||
const claim = await runtime.claimExecutorJob(request.body);
|
||||
return claim ? response.json(claim) : response.status(204).end();
|
||||
} catch (error) {
|
||||
return next(error);
|
||||
}
|
||||
});
|
||||
|
||||
app.post(
|
||||
'/v1/workers/jobs/:jobId/start',
|
||||
requireWorker,
|
||||
async (request, response, next) => {
|
||||
try {
|
||||
return response.json(await runtime.startExecutorJob(request.params.jobId, request.body));
|
||||
} catch (error) {
|
||||
return next(error);
|
||||
}
|
||||
},
|
||||
);
|
||||
|
||||
app.post(
|
||||
'/v1/workers/jobs/:jobId/heartbeat',
|
||||
requireWorker,
|
||||
async (request, response, next) => {
|
||||
try {
|
||||
return response.json(await runtime.heartbeatExecutorJob(request.params.jobId, request.body));
|
||||
} catch (error) {
|
||||
return next(error);
|
||||
}
|
||||
},
|
||||
);
|
||||
|
||||
app.post(
|
||||
'/v1/workers/jobs/:jobId/progress',
|
||||
requireWorker,
|
||||
async (request, response, next) => {
|
||||
try {
|
||||
return response.json(await runtime.progressExecutorJob(request.params.jobId, request.body));
|
||||
} catch (error) {
|
||||
return next(error);
|
||||
}
|
||||
},
|
||||
);
|
||||
|
||||
app.post(
|
||||
'/v1/workers/jobs/:jobId/complete',
|
||||
requireWorker,
|
||||
async (request, response, next) => {
|
||||
try {
|
||||
return response.json(await runtime.completeExecutorJob(request.params.jobId, request.body));
|
||||
} catch (error) {
|
||||
return next(error);
|
||||
}
|
||||
},
|
||||
);
|
||||
|
||||
app.post(
|
||||
'/v1/workers/jobs/:jobId/fail',
|
||||
requireWorker,
|
||||
async (request, response, next) => {
|
||||
try {
|
||||
return response.json(await runtime.failExecutorJob(request.params.jobId, request.body));
|
||||
} catch (error) {
|
||||
return next(error);
|
||||
}
|
||||
},
|
||||
);
|
||||
|
||||
app.post('/v1/workers/recover-expired', requireWorker, async (request, response, next) => {
|
||||
try {
|
||||
return response.json(await runtime.recoverExpiredExecutorJobs(request.body));
|
||||
} catch (error) {
|
||||
return next(error);
|
||||
}
|
||||
});
|
||||
|
||||
app.get('/v1/workers/stats', requireWorker, async (_request, response, next) => {
|
||||
try {
|
||||
return response.json(await runtime.getExecutorQueueStats());
|
||||
} catch (error) {
|
||||
return next(error);
|
||||
}
|
||||
});
|
||||
|
||||
app.post('/v1/executor-jobs/:jobId/cancel', async (request, response, next) => {
|
||||
try {
|
||||
const job = await runtime.cancelExecutorJob(request.params.jobId, request.body);
|
||||
if (!job) {
|
||||
return response.status(404).json({ error: { code: 'EXECUTOR_JOB_NOT_FOUND' } });
|
||||
}
|
||||
return response.json(job);
|
||||
} catch (error) {
|
||||
return next(error);
|
||||
}
|
||||
});
|
||||
|
||||
app.use((error, _request, response, _next) => {
|
||||
response.status(errorStatus(error)).json({
|
||||
error: {
|
||||
@@ -136,5 +294,7 @@ export function createOrchestratorApp({
|
||||
|
||||
export const orchestratorAppInternals = {
|
||||
authorize,
|
||||
authorizeWorker,
|
||||
errorStatus,
|
||||
prometheusMetrics,
|
||||
};
|
||||
|
||||
Reference in New Issue
Block a user