When :replay-required forms throws an error

[kchanqvq]

Currently, it seems that if :replay-required forms throws an error (or non-local exit in general), the workers will be stuck in a positive *replay-forms-counts* and are effective deadlocked.

Can something better than this be done?

[brown]

I'm not sure. I'll give it some thought.

Generally, replay required forms were used to create the initial environment on a set of workers before starting up the actual experiment or parallel computation. The workers are typically running on Borg or Kubernetes, so they may be preempted. If that happens, you want the setup work, the replay required forms, to be evaluated on a new replacement worker before any of the parallel computation is run.

Thanks for reporting this issue. What have you been using swank-crew for? Have you tried the debugging patch for Slime?

[kchanqvq]

Have you tried the debugging patch for Slime?

Do you mean swank.lisp-patch? Yes I applied it.

What have you been using swank-crew for?

I'm using it to add distributed computing support to my research project. Kudos to swank-crew! It's infinitely more productive to have a "distributed" debugger. I'm developing the project just as if it's a local Lisp, which is to say in a very interactive style, so I do generate conditions frequently. I've learnt to always use :replay-required nil and it works very well for now.

Now I understand the intended use-case of :replay-required, I do agree with the assumption that usually error should not happen inside such forms, and the current behavior is acceptable. Handle it more elegantly just improves the robustness of the system a bit.