[Q&A] Flower integration with Flare

[oded-byte]

Python version (python3 -V)

3.12

NVFlare version (python3 -m pip list | grep "nvflare")

2.6.0

NVFlare branch (if running examples, please use the branch that corresponds to the NVFlare version, git branch)

2.6

Operating system

Ubuntu 22.0

Have you successfully run any of the following examples?

• hello-numpy-sag with simulator
• hello-pt with simulator
• hello-numpy-sag with POC
• hello-pt with POC

Please describe your question

Hi everyone,
I am currently exploring the possibility of using an existing Flower (flwr) job and run it using Flare.
I looked at the hello world examples and tried to run it.
Code works perfectly well when just running through flower (flwr run .). The issue is that in client_app of the flower job, they sometime use node_config, rather than run_config:

num_partitions = context.node_config["num-partitions"]
partition_id = context.node_config["partition-id"]

Error I'm getting:
[FLWR-CA@site-2] num_partitions = context.node_config["num-partitions"]
[FLWR-CA@site-2] ~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^
[FLWR-CA@site-2] KeyError: 'num-partitions'

While the run_config can be set using the .toml file, I can't find a way of controlling the node_config file.
How do I populate the node_config file? it is populated when using native flower but unpopulated when using flare.

[chesterxgchen]

@holgerroth can you help with this ?

[holgerroth]

Thanks for raising this issue @oded-byte. I was able to reproduce it using the fl-tabular example from Flower which uses partition_id = context.node_config["partition-id"]. Will investigate.

[holgerroth]

@oded-byte I proposed a PR to fix this #3712

[oded-byte]

Great fix, Thanks!
Should we expect these changes will be available in version 2.7?

[holgerroth]

yes. It is already available in the latest release candidate 2.7.0rc6