Serverless instances die when concurrent

I have a limit of 5 workers. But when I run 3 or so in parallel, often 1-2 of them will die randomly. Doesn't always happen though, not easily reproducible. Is this due to resource constraints?

Anyone else see it? What's the workaround?