Neuron pruning from the subnets is unfair and always favors low UID numbers.

[mutexlocker]

Describe the bug

Some subnets, have almost 90% of their miners with 0 emission. It seems that this logic will always pick the last UID from the UID indexes unless they are immune or have emission. What I am observing is that we are constantly pruning UIDs from the last UIDs, and since we don't have a lot of registrations in the Subnet, this creates a safe haven for UID 0 (or all low UID miners) with 0 emission that will never get pruned and can always try its chance to mine the subnet. We are constantly replacing UIDs between 250 and 255 right now and when they come off immunity then we go beck to 255 again.

To Reproduce

The logic that needs to be fixed is here :

subtensor/pallets/subtensor/src/registration.rs

Line 441 in 1332d07

for neuron_uid_i in 0..neurons_n {

Expected behavior

UID pruning and deregistration in the subnets, when there are a lot of UIDs with zero emissions, should be based on the registration date of the UID or selected randomly among all those that are candidates for pruning.

Screenshots

No response

Environment

Subtensor

Additional context

No response

[VectorChat]

Hey @mutexlocker, I made a PR to attempt to fix this, let me know if it was what you were looking for

[mutexlocker]

for

Thanks for the fix! This was a huge issue in our subnet.

[distributedstatemachine]

closed in #726