wip: pay rewards #607
wip: pay rewards #607
Conversation
akhi3030
force-pushed
the
akhi3030/pay-rewards
branch
5 times, most recently
from
53aebe8 to
c2bf3d0
Compare
akhi3030
force-pushed
the
akhi3030/pay-rewards
branch
from
c2bf3d0 to
817e89b
Compare
| let mut accounts = Arc::unwrap_or_clone(self.vote_accounts()); | ||
| let (_stake, account) = accounts.remove(&voter).unwrap(); | ||
| let data = account.account().data_clone(); | ||
| let mut vote_state: VoteStateV3 = bincode::deserialize(&data).unwrap(); |
There was a problem hiding this comment.
We want to be targetting VoteStateV4 here
| let shared_account = | ||
| AccountSharedData::new_data(account.lamports(), &vote_state, account.owner()).unwrap(); | ||
| // XXX: is the stake getting set properly? | ||
| self.store_account(&voter, &shared_account); |
There was a problem hiding this comment.
Generally prefer store_accounts when storing multiple accounts as AccountsDB is able to use its thread pool to do some of the stuff in parallel.
| // XXX: pay the leader rewards | ||
| let _leader_reward_lamports = reward_lamports - validator_reward_lamports; | ||
|
|
||
| bank.pay_vote_reward(validator, validator_reward_lamports); |
There was a problem hiding this comment.
This isn't quite correct as credits != lamports.
Here's how inflationary rewards are calculated. Let V be the validator set:
Let
points(v) = credits(v) * stake(v) for any validator v
If the total inflation this epoch is T
rewards(v) = T * points(v) / Sum_{u \in V}(points(u))
You can see the full details by poking around in the caller of
There was a problem hiding this comment.
We no longer want the sum of points in the denominator as in Alpenglow rewards are not based on relative performance.
When alpenglow is active we'll want to replace that with 432000 (slots per epoch) * total stake and then I believe you can just add 1 credit here.
There was a problem hiding this comment.
Sorry I realized that the leader also needs to be paid out for the same sol value. This is difficult because the leader's point calculation will multiply by the leader's stake rather than the voter's stake.
I'll need to think more about how to solve the leader's side of rewards.
| } | ||
|
|
||
| pub(crate) fn pay_vote_reward(&self, voter: Pubkey, reward: u64) { | ||
| let mut accounts = Arc::unwrap_or_clone(self.vote_accounts()); |
There was a problem hiding this comment.
Might be able to avoid some of these clones with something like this
let Some((_, vote_account)) = self.vote_accounts().get(&voter) else {
return;
};
let mut vote_state: VoteStateV4 = vote_account.account().deserialize_data();There was a problem hiding this comment.
We could even avoid the deserialize and calculate the offset into epoch credits to update if necessary.
Problem
Summary of Changes
Fixes #