feat(worker): surface child memory state on failures, drop no-op --optimize-for-size

Dockerfile

@@ -108,7 +108,10 @@ COPY package.json README.md LICENSE BUILD.json* ./
 EXPOSE 9090
 # USER node # This would be great to use, but not possible as the volumes are mounted as root
 WORKDIR /app/worker
-CMD ["node", "--disable-warning=ExperimentalWarning", "--optimize-for-size", "index.ts"]
+# Heavy per-run work happens in a child process spawned by the orchestrator (see worker/src/worker.ts).
+# The orchestrator itself only schedules; memory tuning should be done at the container level via
+# NODE_OPTIONS=--max-old-space-size=... and mem_limit, which propagate to both the orchestrator and the child.
+CMD ["node", "--disable-warning=ExperimentalWarning", "index.ts"]
 
 # =============================
 # Install production dependencies for API

tests/features/worker-utils/worker-operations.unit.spec.ts

@@ -1,5 +1,5 @@
 import { test, expect } from '@playwright/test'
-import { buildErrorMessageFromStderr } from '../../../worker/src/utils/worker-operations.ts'
+import { buildErrorMessageFromStderr, formatMemoryUsage, exitCodeHint } from '../../../worker/src/utils/worker-operations.ts'
 
 test.describe('buildErrorMessageFromStderr', () => {
   test('falls back to errMessage when stderr is empty', () => {
@@ -30,3 +30,39 @@ test.describe('buildErrorMessageFromStderr', () => {
     expect(buildErrorMessageFromStderr('a\n\nb\n', 'fb')).toBe('a\nb')
   })
 })
+
+test.describe('formatMemoryUsage', () => {
+  test('renders all components rounded to MB', () => {
+    const mb = 1024 * 1024
+    expect(formatMemoryUsage({
+      rss: 256 * mb,
+      heapUsed: 128 * mb,
+      heapTotal: 200 * mb,
+      external: 16 * mb,
+      arrayBuffers: 0
+    })).toBe('rss=256MB heap=128/200MB ext=16MB')
+  })
+
+  test('returns a string when called without arguments', () => {
+    expect(typeof formatMemoryUsage()).toBe('string')
+  })
+})
+
+test.describe('exitCodeHint', () => {
+  test('returns a V8/SIGABRT hint for code 134', () => {
+    expect(exitCodeHint(134)).toContain('SIGABRT')
+    expect(exitCodeHint(134)).toContain('NODE_OPTIONS')
+  })
+
+  test('returns an OOM-kill hint for code 137', () => {
+    expect(exitCodeHint(137)).toContain('SIGKILL')
+    expect(exitCodeHint(137)).toContain('mem_limit')
+  })
+
+  test('returns empty string for unrelated codes', () => {
+    expect(exitCodeHint(1)).toBe('')
+    expect(exitCodeHint(143)).toBe('')
+    expect(exitCodeHint(null)).toBe('')
+    expect(exitCodeHint(undefined)).toBe('')
+  })
+})

worker/src/task/index.ts

@@ -3,9 +3,14 @@ import nodemailer from 'nodemailer'
 import config from '#config'
 import mongo from '#mongo'
 import { run, stop } from './task.ts'
+import { formatMemoryUsage } from '../utils/worker-operations.ts'
 
 let exitCode = 0
 
+// Memory diagnostic: print on stderr so the parent worker captures it via
+// buildErrorMessageFromStderr when the child exits non-zero.
+console.error(`task start mem ${formatMemoryUsage()}`)
+
 process.on('SIGTERM', function onSigterm () {
   console.info('Received SIGTERM signal, shutdown gracefully...')
   exitCode = 143
@@ -28,4 +33,5 @@ if (err) exitCode = 1
 await mongo.close()
 mailTransport.close()
 
+console.error(`task end mem ${formatMemoryUsage()}`)
 process.exit(exitCode)

worker/src/utils/worker-operations.ts

@@ -18,3 +18,24 @@ export const buildErrorMessageFromStderr = (stderr: string, errMessage: string):
   if (!lines.length) lines.push(errMessage)
   return lines.join('\n')
 }
+
+/**
+ * Format a Node.js MemoryUsage as a compact one-liner, suitable for logging.
+ * All values are rounded to MB.
+ */
+export const formatMemoryUsage = (mem: NodeJS.MemoryUsage = process.memoryUsage()): string => {
+  const mb = (n: number) => Math.round(n / 1024 / 1024)
+  return `rss=${mb(mem.rss)}MB heap=${mb(mem.heapUsed)}/${mb(mem.heapTotal)}MB ext=${mb(mem.external)}MB`
+}
+
+/**
+ * Map a non-zero child exit code to a human hint about likely causes.
+ * Returns an empty string when no specific hint applies.
+ * - 134 = SIGABRT, the signature of a V8 fatal allocation failure (std::bad_alloc / Check failed: (result.ptr) != nullptr).
+ * - 137 = SIGKILL, the signature of an OOM-kill from the host kernel / docker cgroup.
+ */
+export const exitCodeHint = (code: number | null | undefined): string => {
+  if (code === 134) return 'le processus enfant a abandonné (SIGABRT, code 134) — typique d\'une allocation V8 impossible. Vérifier NODE_OPTIONS=--max-old-space-size et la limite mémoire du conteneur (mem_limit / resources.limits.memory).'
+  if (code === 137) return 'le processus enfant a été tué (SIGKILL, code 137) — typique d\'un OOM-kill par le noyau / cgroup. Augmenter la limite mémoire du conteneur (mem_limit / resources.limits.memory).'
+  return ''
+}

worker/src/worker.ts

@@ -18,7 +18,7 @@ import locks from '#locks'
 import limits from './utils/limits.ts'
 import { initMetrics } from './utils/metrics.ts'
 import { finish } from './utils/runs.ts'
-import { buildErrorMessageFromStderr } from './utils/worker-operations.ts'
+import { buildErrorMessageFromStderr, exitCodeHint } from './utils/worker-operations.ts'
 
 const debug = Debug('worker')
 const debugLoop = Debug('worker-loop')
@@ -242,8 +242,11 @@ async function iter (run: Run) {
     })
     await finish(run)
   } catch (err: any) {
-    // Build back the original error message from the stderr of the child process
-    const errorMessage = buildErrorMessageFromStderr(stderr, err.message)
+    // Build back the original error message from the stderr of the child process,
+    // appending a hint when the child exit code matches a known OOM signature.
+    const baseMessage = buildErrorMessageFromStderr(stderr, err.message)
+    const hint = exitCodeHint(err.code)
+    const errorMessage = hint ? `${baseMessage}\n${hint}` : baseMessage
 
     if (run) {
       // case of interruption by a SIGTERM