Is there an existing proposal for this?

• I have searched the existing proposals

Is your feature request related to a problem?

The cloud provider we use uses Flatcar Container Linux for its Kubernetes nodes. They compile their kernel with CONFIG_CROSS_MEMORY_ATTACH not set, which means process_vm_readv is not compiled in.

This means, if we try to use pystack remote <pid> from a kubectl debug container we see 💀 Engine error: Function not implemented 💀.

Describe the solution you'd like

I think it could make sense to fall back to reading the memory through /proc/<pid>/mem if process_vm_readv returns ENOSYS.

I should probably confess, I found this when testing one of my own tools, and was now testing pystack out of curiosity. I used pystack in the past though, before we switched cloud provider, and it would be a shame not to be able to call on it again when we need it.

This idea of using /proc/<pid>/mem is what I've done in pymontrace, cakemanny/pymontrace#1 , based on what I read that py-spy has done, so I can confirm it works (as a fallback for process_vm_writev in my case).

I think the only real drawbacks are

• it causes between 2 and 4 extra syscalls per read (depending on how naive the implementation is)
• the memory has to be transferred through the kernel
  but it's only a penalty for those with such a kernel.

Alternatives you considered

No response