Igor Davidyuk

Corrupt shard utils and experiments

1

This PR introduces Corrupting Shard Descriptor wrapper. Decorating a Shard Descriptor (SD) class with this function will result in the following: 1. `corrupt` parameter is added to the SD init...

Introduce a deterministic algorithm for choosing a port for the aggregator server.

1

The workflow will probably be as follows: Director config will feature a new parameter `open_ports` listing one or more ports. This list will be tracked by the Director-service, which will...

1. Default opt_treatment value causing an error https://github.com/intel/openfl/blob/bff461a20c535cb75eebdbc6ace2895b5b560f02/openfl/component/collaborator/collaborator.py#L70 2. Interactive API's Experiment object should operate in enum values for opt_treatment parameter and device_assignment_policy https://github.com/intel/openfl/blob/bff461a20c535cb75eebdbc6ace2895b5b560f02/openfl/interface/interactive_api/experiment.py#L149 3. TaskRunners (including the old API)...

This line may suppress traceback for exceptions. https://github.com/intel/openfl/blob/aed8452b12d20dcc80bd722336246d679fa41484/openfl/component/envoy/envoy.py#L68 Regardless, OpenFL could do better by forwarding exceptions and tracebacks from aggregator and collaborator to the Frontend client

I faced errors running `torch_unet_kvasir` and `torch_unet_kvasir_gramine_ready` examples. The message is attached [error.log](https://github.com/intel/openfl/files/9597992/error.log)

WIP: Testing communication resilience

1

This PR introduces a bash script that emulates cutting a network connection of one or several FL actors. All the actors run in docker containers and are connected to a...

Lighter text encoder

2

Hey guys, thank you for the awesome repo! Is there any chance that you will support a lighter tex encoder?

Elaborate docs on TaskRunner interface

1

We have the TaskRunner interface is scarcely described in docs, adding details would make it more transparent for users. Specifically, I received questions regarding `FL task` methods signatures.

Training fails with the following message: `DefaultCPUAllocator: can't allocate memory: you tried to allocate 225738752 bytes. Error code 12 (Cannot allocate memory)` on the collaborator side. [Logs attached](https://github.com/intel/openfl/files/9951006/gramine-kvasir.log)

**Describe the bug** Envoy's collaborator may infinitely try reconnecting to the aggregator, while it had already shut down by Director service. **To Reproduce** Steps to reproduce the behavior: 1. Enable...