#19769: Chrome tab crashes ("Aw, Snap!") on self-hosted Twenty v1.21.2

Describe the bug

Chrome tabs crash with "Aw, Snap!" (error code 5) when navigating a self-hosted Twenty instance. The crashes are random but reproducible. Server logs show zero errors during crashes, suggesting this is a frontend/browser memory issue rather than a backend problem.

Environment

Reproduction

The crash occurs randomly during normal navigation. Three patterns make it more likely:

1. Switching between workspaces rapidly in the same tab
2. Long AI chat conversations (the accumulated context seems to grow until the tab runs out of memory)
3. Opening the Settings page (which loads a significant amount of metadata)

Sometimes the tab crashes within seconds of loading; sometimes it takes a few minutes.

What we have tried

• Upgraded the droplet from 4GB to 8GB RAM and added 4GB swap
• Set NODE_OPTIONS=--max-old-space-size=4096 on both server and worker containers
• Upgraded incrementally from v1.19 through v1.20 to v1.21.2
• Cleared browser cookies and cache
• Tested in incognito mode (still crashes)
• Configured Caddy with flush_interval -1 for SSE streaming
• Tested with direct port access (bypassing Caddy entirely)

What we have ruled out

• Server resources: 6GB+ RAM free during crashes, CPU idle, all containers healthy
• Network: healthz returns 200 in under 100ms
• Browser state: crashes happen in incognito with no extensions
• Reverse proxy: crashes also occurred when hitting the server directly

Questions

Has anyone else experienced Chrome tab crashes on a self-hosted instance? We noticed PR #19213 references a memory leak in v1.21. Could this be related?

Are there any recommended frontend performance settings or configurations for self-hosted deployments that we might be missing?

Happy to provide browser memory profiles, HAR files, or Docker logs if that would help narrow this down.

Twenty version: v1.21.2 Self-hosted: Yes (Docker)