search

ray-project / rayfed

A multiple parties joint, distributed execution engine based on Ray, to help build your own federated learning frameworks in minutes.
https://rayfed.readthedocs.io
Apache License 2.0
92 stars 22 forks source link

clear global_context after shutdown #114

Closed NKcqx closed 1 year ago

NKcqx commented 1 year ago

The global_context incrementally generates the fed task id. It should be reset after calling fed.shutdown so that the fed task id can grow from 0 again.

NKcqx commented 1 year ago

One potential risk is that in multiple controller mode, we don't barrier the controll flow at the node fed.shutdown, so party A may send a fresh data from the new session, but another party is still receiving the old data.

Indeed, I've considered this case, will file a new issue and mention it in the next coming barrier PR.

jovany-wang commented 1 year ago

Indeed, I've considered this case, will file a new issue and mention it in the next coming barrier PR.

Great! Maybe you could take a look at this one https://github.com/ray-project/rayfed/issues/66 , which is similar.