Gateway startup takes 4-10 minutes (603 JS chunks, 2.2GB memory peak)

Summary

Gateway takes ~4 minutes on warm restart and ~10 minutes on cold boot before it begins accepting connections. During startup, Node.js consumes up to 2.2GB of memory and produces no log output — it appears to be parsing/compiling the JS bundle.

Environment

• OpenClaw version: 2026.3.14
• Node.js: v22.22.0
• OS: Ubuntu 24.04 VM, Linux 6.17, x86_64
• Resources: 4 vCPU, 32GB RAM, SSD-backed storage
• Install method: source (git clone + pnpm install + pnpm build)
• Service manager: systemd (user unit)

Bundle stats

dist/: 603 JS files, 69MB total
node_modules/: 2.5GB

Workload

• 8 cron jobs
• 5 channel sessions (Discord + WhatsApp)
• 57 total sessions (1.5MB sessions.json)
• Browser relay enabled

Observed behavior

Warm restart (disk cache hot)

18:58:09  systemd starts node process
19:02:14  first gateway log line (canvas mounted, heartbeat started)

~4 minutes of silence between process start and first output.

Cold boot (after VM reboot)

18:18:10  systemd starts node process
18:28:15  first gateway log line

~10 minutes of silence.

Memory

• Peak memory during startup: 2.2GB (reported by systemd MemoryPeak)
• At default Node.js heap limit (~1.7GB), the gateway OOM crashes during startup or under load:

  FATAL ERROR: Reached heap limit Allocation failed - JavaScript heap out of memory
  

• The OOM occurs during JSON.parse in the V8 stack trace
• Workaround: --max-old-space-size=4096

Systemd restart counter

The service has OOM-crashed and auto-restarted 5 times over ~2 weeks of operation.

Expected behavior

Gateway should start in under 30 seconds. 603 separate JS chunks seems high — a single bundled entry point would dramatically reduce module resolution and V8 compilation overhead.

Workaround

Setting --max-old-space-size=4096 in the systemd service prevents OOM crashes but doesn't improve startup time.

Questions

1. Is the 603-chunk dist layout intentional (code splitting for lazy loading)?
2. Are there plans for a single-file bundle or V8 snapshot to improve cold start?
3. Is 2.2GB memory at startup expected, or could there be a memory leak during initialization?

Additional context

• Once running, the gateway performs well and memory stabilizes
• Warm restarts (SIGUSR1 config reload) within a running process are fast — it's only full process restarts that are slow
• The session state is small (57 sessions, 1.5MB) so this doesn't appear to be data-related