1import { loadDotenv } from './lib/env.ts';2loadDotenv();3process.env.RI_SERVICE ??= 'rareindex-worker';45import { env, logger } from '@rareindex/shared';6import { createQueue, JOBS } from './lib/queue.ts';7import { closeDb, db } from './lib/db.ts';8import { flushCosts } from './lib/costs.ts';9import { runCrawl } from './crawler/run.ts';10import { scheduleDueCrawls } from './crawler/scheduler.ts';11import { normalizeBatch } from './normalizer/index.ts';12import { sql } from 'drizzle-orm';13import { pendingCount, resolveBatch } from './entity-resolution/index.ts';14import { assetsNeedingValuation, computePremiums, valueMany } from './valuation/run.ts';15import { runCategorySnapshots, runIndices, runRadar } from './indices/run.ts';16import { syncFx } from './fx.ts';17import { syncBenchmarks } from './benchmarks.ts';18import { computeHealth } from './health.ts';19import { expireListings } from './listings-expire.ts';20import { processImages } from './image-processing/index.ts';21import { verifyCertificates } from './certs-verify.ts';2223/**24 * RareIndex worker process (PM2: rareindex-worker). One process runs the queue handlers and the25 * scheduler; scale out by starting more processes against the same database.26 */27export async function startWorker(): Promise<() => Promise<void>> {28 const log = logger.child({ component: 'worker' });29 const queue = createQueue();30 await queue.start();31 // Single-worker deployment: jobs left 'active' by a previous process (restart/crash) would wait for32 // their expiry (hours). Re-queue them so crawls resume immediately.33 await db().execute(sql`update pgboss.job set state = 'retry', started_on = null where state = 'active' and started_on < now() - interval '90 seconds'`);34 // Runs left 'running' by a dead process can never finish: close them so the per-connector guard below works.35 await db().execute(sql`update connector_runs set status = 'failed', finished_at = now(), error = 'worker restarted' where status = 'running'`);36 // Collapse duplicate queued crawls per connector (keep the oldest).37 await db().execute(sql`delete from pgboss.job j using pgboss.job k where j.name = 'crawl.run' and k.name = 'crawl.run' and j.state in ('created','retry') and k.state in ('created','retry') and j.data->>'connectorId' = k.data->>'connectorId' and j.created_on > k.created_on`);38 const concurrency = env().WORKER_CONCURRENCY;3940 // ---- handlers ----41 await queue.work<{ connectorId: string; mode?: 'incremental' | 'backfill' | 'probe'; limit?: number; trigger?: string }>(JOBS.crawlRun, { concurrency: Math.max(1, Math.floor(concurrency / 2)), pollingIntervalSeconds: 5 }, async (data) => {42 // One crawl per connector at a time (duplicates waste engine credits and crawl slots).43 const [running] = (await db().execute(sql`select 1 from connector_runs where connector_id = ${data.connectorId} and status = 'running' and started_at > now() - interval '6 hours' limit 1`)) as unknown as unknown[];44 if (running) {45 log.info({ connector: data.connectorId }, 'crawl already running; skipping duplicate job');46 return;47 }48 const res = await runCrawl(data.connectorId, { mode: data.mode ?? 'incremental', limit: data.limit, trigger: data.trigger ?? 'schedule' });49 if (res.recordsRaw > 0) await queue.send(JOBS.normalizeBatch, { connectorId: data.connectorId }, { singletonKey: `normalize:${data.connectorId}`, singletonSeconds: 30 });50 });51 await queue.work<{ connectorId?: string }>(JOBS.normalizeBatch, { concurrency: 2, pollingIntervalSeconds: 5 }, async (data) => {52 let total = 0;53 let more = false;54 for (let i = 0; i < 40; i++) {55 const r = await normalizeBatch({ connectorId: data.connectorId, limit: 500 });56 total += r.processed;57 more = r.processed >= 500;58 if (!more) break;59 }60 // Large crawls (100k+ raw rows) exceed one job's budget: chain another job instead of leaving rows unprocessed.61 if (more) await queue.send(JOBS.normalizeBatch, { connectorId: data.connectorId }, { singletonKey: `normalize:${data.connectorId ?? 'all'}:${Date.now()}`, startAfterSeconds: 1 });62 if (total > 0) for (let k = 0; k < Number(process.env.RESOLVE_CONCURRENCY ?? 3); k++) await queue.send(JOBS.resolveBatch, {}, { singletonKey: `resolve:${k}`, singletonSeconds: 30 });63 });64 const RESOLVE_WORKERS = Number(process.env.RESOLVE_CONCURRENCY ?? 3);65 // Release rows claimed by a previous worker process that died mid-batch.66 await db().execute(sql`update normalized_records set status = 'pending' where status = 'processing'`);67 await queue.work<Record<string, never>>(JOBS.resolveBatch, { concurrency: RESOLVE_WORKERS, pollingIntervalSeconds: 5 }, async () => {68 const touched = new Set<string>();69 for (let i = 0; i < 40; i++) {70 const r = await resolveBatch({ limit: 500 });71 for (const a of r.touchedAssets) touched.add(a);72 if (r.processed < 500) break;73 }74 if ((await pendingCount()) > 0) for (let k = 0; k < RESOLVE_WORKERS; k++) await queue.send(JOBS.resolveBatch, {}, { singletonKey: `resolve:${k}`, singletonSeconds: 30, startAfterSeconds: 5 });75 if (touched.size) await queue.send(JOBS.valuationAsset, { assetIds: [...touched].slice(0, 5000) }, { singletonKey: `value:${Date.now()}` });76 if (touched.size) await queue.send(JOBS.imagesProcess, { assetIds: [...touched].slice(0, 5000), limit: 5000 }, { singletonKey: `images:${Date.now()}` });77 });78 await queue.work<{ assetIds: string[] }>(JOBS.valuationAsset, { concurrency: 1, pollingIntervalSeconds: 5 }, async (data) => {79 await valueMany(data.assetIds, { concurrency: 4 });80 });81 await queue.work<{ all?: boolean; rebuildHistory?: boolean }>(JOBS.valuationRebuild, { concurrency: 1, pollingIntervalSeconds: 30 }, async (data) => {82 await computePremiums();83 const ids = await assetsNeedingValuation({ all: data.all, limit: 200_000 });84 await valueMany(ids, { rebuildHistory: data.rebuildHistory, concurrency: 4 });85 });86 await queue.work(JOBS.indicesDaily, { concurrency: 1, pollingIntervalSeconds: 30 }, async () => {87 await runIndices();88 });89 await queue.work(JOBS.snapshotsDaily, { concurrency: 1, pollingIntervalSeconds: 30 }, async () => {90 await runCategorySnapshots();91 });92 await queue.work(JOBS.radarScan, { concurrency: 1, pollingIntervalSeconds: 30 }, async () => {93 await runRadar();94 });95 await queue.work(JOBS.healthCompute, { concurrency: 1, pollingIntervalSeconds: 30 }, async () => {96 await computeHealth();97 });98 await queue.work<{ backfill?: boolean }>(JOBS.fxSync, { concurrency: 1, pollingIntervalSeconds: 30 }, async (data) => {99 await syncFx({ backfill: data.backfill });100 });101 await queue.work(JOBS.benchmarksSync, { concurrency: 1, pollingIntervalSeconds: 60 }, async () => {102 await syncBenchmarks();103 });104 await queue.work<{ assetIds?: string[]; limit?: number; recheck?: boolean }>(JOBS.imagesProcess, { concurrency: 1, pollingIntervalSeconds: 30 }, async (data) => {105 await processImages({ assetIds: data.assetIds, limit: data.limit ?? 2000, recheck: data.recheck ?? false });106 });107 await queue.work(JOBS.listingsExpire, { concurrency: 1, pollingIntervalSeconds: 60 }, async () => {108 await expireListings();109 });110 await queue.work<{ limit?: number; grader?: string }>(JOBS.certsVerify, { concurrency: 1, pollingIntervalSeconds: 60 }, async (data) => {111 await verifyCertificates({ limit: data.limit ?? 100, grader: data.grader });112 });113 await queue.work(JOBS.accountJobs, { concurrency: 1, pollingIntervalSeconds: 60 }, async () => {114 try {115 const accountModule = './account/index.ts'; // provided by the account module when present116 const mod = (await import(accountModule)) as { runAccountJobs?: (d: ReturnType<typeof db>) => Promise<unknown> };117 if (mod.runAccountJobs) await mod.runAccountJobs(db());118 } catch (err) {119 if (!/Cannot find module|ERR_MODULE_NOT_FOUND/.test(String(err))) log.error({ err }, 'account jobs failed');120 }121 });122123 // ---- schedules (UTC) ----124 await queue.schedule(JOBS.fxSync, '15 16 * * 1-5', {}); // after ECB 16:00 CET publication125 await queue.schedule(JOBS.benchmarksSync, '30 22 * * *', {});126 await queue.schedule(JOBS.valuationRebuild, '0 3 * * *', { all: false });127 await queue.schedule(JOBS.snapshotsDaily, '20 4 * * *', {});128 await queue.schedule(JOBS.indicesDaily, '40 4 * * *', {});129 await queue.schedule(JOBS.radarScan, '0 5 * * *', {});130 await queue.schedule(JOBS.healthCompute, '*/30 * * * *', {});131 await queue.schedule(JOBS.listingsExpire, '10 * * * *', {});132 await queue.schedule(JOBS.certsVerify, '25 * * * *', { limit: 100 }); // SPEC §22: verify cert numbers seen on marketplaces against public grader lookups133 await queue.schedule(JOBS.imagesProcess, '50 5 * * *', { limit: 20000, recheck: true });134 await queue.schedule(JOBS.imagesProcess, '*/15 * * * *', { limit: 3000 });135 await queue.schedule(JOBS.accountJobs, '*/10 * * * *', {});136137 // ---- crawl scheduler loop ----138 const tick = async () => {139 try {140 await scheduleDueCrawls(queue);141 // Normalize whatever raw rows exist even while long crawls are still running (sales show up progressively).142 await queue.send(JOBS.normalizeBatch, {}, { singletonKey: 'normalize:periodic', singletonSeconds: 120 });143 } catch (err) {144 log.error({ err }, 'scheduler tick failed');145 }146 };147 await tick();148 if ((await pendingCount()) > 0) for (let k = 0; k < RESOLVE_WORKERS; k++) await queue.send(JOBS.resolveBatch, {}, { singletonKey: `resolve:${k}`, singletonSeconds: 30 });149 const timer = setInterval(tick, 60_000);150 // hourly incremental valuation for assets with fresh evidence151 const hourly = setInterval(() => void queue.send(JOBS.valuationRebuild, { all: false }, { singletonKey: 'valuation:hourly', singletonSeconds: 3000 }), 3600_000);152 log.info({ concurrency }, 'worker started');153154 return async () => {155 clearInterval(timer);156 clearInterval(hourly);157 await queue.stop();158 await flushCosts();159 await closeDb();160 log.info('worker stopped');161 };162}163164const isMain = process.argv[1] && /workers\/main\.ts$/.test(process.argv[1]);165if (isMain) {166 startWorker()167 .then((stop) => {168 const shutdown = () => void stop().then(() => process.exit(0));169 process.on('SIGINT', shutdown);170 process.on('SIGTERM', shutdown);171 })172 .catch((err) => {173 logger.error({ err }, 'worker failed to start');174 process.exit(1);175 });176}177