Hi everyone,

I'm running a website audit service on Railway that uses Playwright/Chromium to perform automated cookie consent audits on websites.

**My Setup:**
- Service: Node.js 20 backend
- Playwright with Chromium
- Browser Pool: 4 browsers × 10 contexts = 40 parallel browser contexts
- Resources: 32 vCPU, 32GB RAM (plenty available)
- Environment: Railway staging

SHM_SIZE_BYTES=1073741824 (1GB, also tried 4GB) via Railway specific env-variable

My Browser headers are: 

const args: string[] = [
      '--disable-http2',
      '--disable-gpu',
      '--allow-insecure-localhost',
    ];


**The Problem:**
When I try to start 40 browser contexts simultaneously, the browser processes crash immediately with errors like:
- `Target crashed`
- `Browser unexpectedly closed`
- `page.evaluate: Target crashed`

Peak Usage was: 6GB Ram and 7.9vCPU
This happens **even though CPU and RAM are barely utilized**

**What Works:**
Interestingly, a smaller configuration runs perfectly fine:
- 2 browsers × 12 contexts = 24 parallel browser contexts
- 24 contexts total
- No crashes, stable performance

Peak-Usage: 13vCPU and 6GB RAM

**What I've Already Checked:**
- CPU/RAM are not the bottleneck
- Increased file descriptors (ulimit -n 65536)
- Increased process limits (ulimit -u 4096)
- No "EMFILE: too many open files" errors
- No memory allocation failures
- Removing the '--disable-http2' browser header
- Adding the '--disable-dev-shm-usage' browser header

**My Question:**
Are there any **non-obvious or undocumented limits on Railway** that could cause this? For example:
- Kernel-level process spawning limits?
- Container isolation limits?
- Network connection limits?
- Other resource constraints not visible via `top` or `free`?

The jump from 24 contexts (stable) to 40 contexts (crashes) suggests there's a hard limit somewhere that I'm not seeing.

Any insights would be greatly appreciated!

Thanks!