Batch worker cannot start with more than 64 CPUs

[cjllanwarne]

What happened?

Try running a job with _machine_type: 'n1-highmem-64'. This is necessary to get enough memory for some larger jobs (> ~200GB).

Startup on the batch worker fails because the job is calculating how many theoretical network namespaces it could support (4 per CPU, 64 CPUS, plus some for JVMs), but not considering that the IPv4 schema puts a hard limit of 255 on namespaces if only one subnet value is changing each time.

Version

Live 7/30/24

Relevant log output

No response

Security considerations:

Low risk of impacting security. High CPU machine types are not materially different from others with respect to security considerations, and the bug is a simple logic error.