lighthouse/beacon_node/beacon_chain/tests/rewards.rs

682 lines
24 KiB
Rust
Raw Normal View History

#![cfg(test)]
use std::collections::HashMap;
use beacon_chain::test_utils::{
generate_deterministic_keypairs, BeaconChainHarness, EphemeralHarnessType,
};
use beacon_chain::{
test_utils::{AttestationStrategy, BlockStrategy, RelativeSyncCommittee},
types::{Epoch, EthSpec, Keypair, MinimalEthSpec},
};
Phase 0 attestation rewards via Beacon API (#4474) ## Issue Addressed Addresses #4026. Beacon-API spec [here](https://ethereum.github.io/beacon-APIs/?urls.primaryName=dev#/Beacon/getAttestationsRewards). Endpoint: `POST /eth/v1/beacon/rewards/attestations/{epoch}` This endpoint already supports post-Altair epochs. This PR adds support for phase 0 rewards calculation. ## Proposed Changes - [x] Attestation rewards API to support phase 0 rewards calculation, re-using logic from `state_processing`. Refactored `get_attestation_deltas` slightly to support computing deltas for a subset of validators. - [x] Add `inclusion_delay` to `ideal_rewards` (`beacon-API` spec update to follow) - [x] Add `inactivity` penalties to both `ideal_rewards` and `total_rewards` (`beacon-API` spec update to follow) - [x] Add tests to compute attestation rewards and compare results with beacon states ## Additional Notes - The extra penalty for missing attestations or being slashed during an inactivity leak is currently not included in the API response (for both phase 0 and Altair) in the spec. - I went with adding `inactivity` as a separate component rather than combining them with the 4 rewards, because this is how it was grouped in [the phase 0 spec](https://github.com/ethereum/consensus-specs/blob/dev/specs/phase0/beacon-chain.md#get_attestation_deltas). During inactivity leak, all rewards include the optimal reward, and inactivity penalties are calculated separately (see below code snippet from the spec), so it would be quite confusing if we merge them. This would also work better with Altair, because there's no "cancelling" of rewards and inactivity penalties are more separate. - Altair calculation logic (to include inactivity penalties) to be updated in a follow-up PR. ```python def get_attestation_deltas(state: BeaconState) -> Tuple[Sequence[Gwei], Sequence[Gwei]]: """ Return attestation reward/penalty deltas for each validator. """ source_rewards, source_penalties = get_source_deltas(state) target_rewards, target_penalties = get_target_deltas(state) head_rewards, head_penalties = get_head_deltas(state) inclusion_delay_rewards, _ = get_inclusion_delay_deltas(state) _, inactivity_penalties = get_inactivity_penalty_deltas(state) rewards = [ source_rewards[i] + target_rewards[i] + head_rewards[i] + inclusion_delay_rewards[i] for i in range(len(state.validators)) ] penalties = [ source_penalties[i] + target_penalties[i] + head_penalties[i] + inactivity_penalties[i] for i in range(len(state.validators)) ] return rewards, penalties ``` ## Example API Response <details> <summary>Click me</summary> ```json { "ideal_rewards": [ { "effective_balance": "1000000000", "head": "6638", "target": "6638", "source": "6638", "inclusion_delay": "9783", "inactivity": "0" }, { "effective_balance": "2000000000", "head": "13276", "target": "13276", "source": "13276", "inclusion_delay": "19565", "inactivity": "0" }, { "effective_balance": "3000000000", "head": "19914", "target": "19914", "source": "19914", "inclusion_delay": "29349", "inactivity": "0" }, { "effective_balance": "4000000000", "head": "26553", "target": "26553", "source": "26553", "inclusion_delay": "39131", "inactivity": "0" }, { "effective_balance": "5000000000", "head": "33191", "target": "33191", "source": "33191", "inclusion_delay": "48914", "inactivity": "0" }, { "effective_balance": "6000000000", "head": "39829", "target": "39829", "source": "39829", "inclusion_delay": "58697", "inactivity": "0" }, { "effective_balance": "7000000000", "head": "46468", "target": "46468", "source": "46468", "inclusion_delay": "68480", "inactivity": "0" }, { "effective_balance": "8000000000", "head": "53106", "target": "53106", "source": "53106", "inclusion_delay": "78262", "inactivity": "0" }, { "effective_balance": "9000000000", "head": "59744", "target": "59744", "source": "59744", "inclusion_delay": "88046", "inactivity": "0" }, { "effective_balance": "10000000000", "head": "66383", "target": "66383", "source": "66383", "inclusion_delay": "97828", "inactivity": "0" }, { "effective_balance": "11000000000", "head": "73021", "target": "73021", "source": "73021", "inclusion_delay": "107611", "inactivity": "0" }, { "effective_balance": "12000000000", "head": "79659", "target": "79659", "source": "79659", "inclusion_delay": "117394", "inactivity": "0" }, { "effective_balance": "13000000000", "head": "86298", "target": "86298", "source": "86298", "inclusion_delay": "127176", "inactivity": "0" }, { "effective_balance": "14000000000", "head": "92936", "target": "92936", "source": "92936", "inclusion_delay": "136959", "inactivity": "0" }, { "effective_balance": "15000000000", "head": "99574", "target": "99574", "source": "99574", "inclusion_delay": "146742", "inactivity": "0" }, { "effective_balance": "16000000000", "head": "106212", "target": "106212", "source": "106212", "inclusion_delay": "156525", "inactivity": "0" }, { "effective_balance": "17000000000", "head": "112851", "target": "112851", "source": "112851", "inclusion_delay": "166307", "inactivity": "0" }, { "effective_balance": "18000000000", "head": "119489", "target": "119489", "source": "119489", "inclusion_delay": "176091", "inactivity": "0" }, { "effective_balance": "19000000000", "head": "126127", "target": "126127", "source": "126127", "inclusion_delay": "185873", "inactivity": "0" }, { "effective_balance": "20000000000", "head": "132766", "target": "132766", "source": "132766", "inclusion_delay": "195656", "inactivity": "0" }, { "effective_balance": "21000000000", "head": "139404", "target": "139404", "source": "139404", "inclusion_delay": "205439", "inactivity": "0" }, { "effective_balance": "22000000000", "head": "146042", "target": "146042", "source": "146042", "inclusion_delay": "215222", "inactivity": "0" }, { "effective_balance": "23000000000", "head": "152681", "target": "152681", "source": "152681", "inclusion_delay": "225004", "inactivity": "0" }, { "effective_balance": "24000000000", "head": "159319", "target": "159319", "source": "159319", "inclusion_delay": "234787", "inactivity": "0" }, { "effective_balance": "25000000000", "head": "165957", "target": "165957", "source": "165957", "inclusion_delay": "244570", "inactivity": "0" }, { "effective_balance": "26000000000", "head": "172596", "target": "172596", "source": "172596", "inclusion_delay": "254352", "inactivity": "0" }, { "effective_balance": "27000000000", "head": "179234", "target": "179234", "source": "179234", "inclusion_delay": "264136", "inactivity": "0" }, { "effective_balance": "28000000000", "head": "185872", "target": "185872", "source": "185872", "inclusion_delay": "273918", "inactivity": "0" }, { "effective_balance": "29000000000", "head": "192510", "target": "192510", "source": "192510", "inclusion_delay": "283701", "inactivity": "0" }, { "effective_balance": "30000000000", "head": "199149", "target": "199149", "source": "199149", "inclusion_delay": "293484", "inactivity": "0" }, { "effective_balance": "31000000000", "head": "205787", "target": "205787", "source": "205787", "inclusion_delay": "303267", "inactivity": "0" }, { "effective_balance": "32000000000", "head": "212426", "target": "212426", "source": "212426", "inclusion_delay": "313050", "inactivity": "0" } ], "total_rewards": [ { "validator_index": "0", "head": "212426", "target": "212426", "source": "212426", "inclusion_delay": "313050", "inactivity": "0" }, { "validator_index": "32", "head": "212426", "target": "212426", "source": "212426", "inclusion_delay": "313050", "inactivity": "0" }, { "validator_index": "63", "head": "-357771", "target": "-357771", "source": "-357771", "inclusion_delay": "0", "inactivity": "0" } ] } ``` </details>
2023-07-18 01:48:40 +00:00
use eth2::lighthouse::attestation_rewards::TotalAttestationRewards;
use eth2::lighthouse::StandardAttestationRewards;
use eth2::types::ValidatorId;
use lazy_static::lazy_static;
Phase 0 attestation rewards via Beacon API (#4474) ## Issue Addressed Addresses #4026. Beacon-API spec [here](https://ethereum.github.io/beacon-APIs/?urls.primaryName=dev#/Beacon/getAttestationsRewards). Endpoint: `POST /eth/v1/beacon/rewards/attestations/{epoch}` This endpoint already supports post-Altair epochs. This PR adds support for phase 0 rewards calculation. ## Proposed Changes - [x] Attestation rewards API to support phase 0 rewards calculation, re-using logic from `state_processing`. Refactored `get_attestation_deltas` slightly to support computing deltas for a subset of validators. - [x] Add `inclusion_delay` to `ideal_rewards` (`beacon-API` spec update to follow) - [x] Add `inactivity` penalties to both `ideal_rewards` and `total_rewards` (`beacon-API` spec update to follow) - [x] Add tests to compute attestation rewards and compare results with beacon states ## Additional Notes - The extra penalty for missing attestations or being slashed during an inactivity leak is currently not included in the API response (for both phase 0 and Altair) in the spec. - I went with adding `inactivity` as a separate component rather than combining them with the 4 rewards, because this is how it was grouped in [the phase 0 spec](https://github.com/ethereum/consensus-specs/blob/dev/specs/phase0/beacon-chain.md#get_attestation_deltas). During inactivity leak, all rewards include the optimal reward, and inactivity penalties are calculated separately (see below code snippet from the spec), so it would be quite confusing if we merge them. This would also work better with Altair, because there's no "cancelling" of rewards and inactivity penalties are more separate. - Altair calculation logic (to include inactivity penalties) to be updated in a follow-up PR. ```python def get_attestation_deltas(state: BeaconState) -> Tuple[Sequence[Gwei], Sequence[Gwei]]: """ Return attestation reward/penalty deltas for each validator. """ source_rewards, source_penalties = get_source_deltas(state) target_rewards, target_penalties = get_target_deltas(state) head_rewards, head_penalties = get_head_deltas(state) inclusion_delay_rewards, _ = get_inclusion_delay_deltas(state) _, inactivity_penalties = get_inactivity_penalty_deltas(state) rewards = [ source_rewards[i] + target_rewards[i] + head_rewards[i] + inclusion_delay_rewards[i] for i in range(len(state.validators)) ] penalties = [ source_penalties[i] + target_penalties[i] + head_penalties[i] + inactivity_penalties[i] for i in range(len(state.validators)) ] return rewards, penalties ``` ## Example API Response <details> <summary>Click me</summary> ```json { "ideal_rewards": [ { "effective_balance": "1000000000", "head": "6638", "target": "6638", "source": "6638", "inclusion_delay": "9783", "inactivity": "0" }, { "effective_balance": "2000000000", "head": "13276", "target": "13276", "source": "13276", "inclusion_delay": "19565", "inactivity": "0" }, { "effective_balance": "3000000000", "head": "19914", "target": "19914", "source": "19914", "inclusion_delay": "29349", "inactivity": "0" }, { "effective_balance": "4000000000", "head": "26553", "target": "26553", "source": "26553", "inclusion_delay": "39131", "inactivity": "0" }, { "effective_balance": "5000000000", "head": "33191", "target": "33191", "source": "33191", "inclusion_delay": "48914", "inactivity": "0" }, { "effective_balance": "6000000000", "head": "39829", "target": "39829", "source": "39829", "inclusion_delay": "58697", "inactivity": "0" }, { "effective_balance": "7000000000", "head": "46468", "target": "46468", "source": "46468", "inclusion_delay": "68480", "inactivity": "0" }, { "effective_balance": "8000000000", "head": "53106", "target": "53106", "source": "53106", "inclusion_delay": "78262", "inactivity": "0" }, { "effective_balance": "9000000000", "head": "59744", "target": "59744", "source": "59744", "inclusion_delay": "88046", "inactivity": "0" }, { "effective_balance": "10000000000", "head": "66383", "target": "66383", "source": "66383", "inclusion_delay": "97828", "inactivity": "0" }, { "effective_balance": "11000000000", "head": "73021", "target": "73021", "source": "73021", "inclusion_delay": "107611", "inactivity": "0" }, { "effective_balance": "12000000000", "head": "79659", "target": "79659", "source": "79659", "inclusion_delay": "117394", "inactivity": "0" }, { "effective_balance": "13000000000", "head": "86298", "target": "86298", "source": "86298", "inclusion_delay": "127176", "inactivity": "0" }, { "effective_balance": "14000000000", "head": "92936", "target": "92936", "source": "92936", "inclusion_delay": "136959", "inactivity": "0" }, { "effective_balance": "15000000000", "head": "99574", "target": "99574", "source": "99574", "inclusion_delay": "146742", "inactivity": "0" }, { "effective_balance": "16000000000", "head": "106212", "target": "106212", "source": "106212", "inclusion_delay": "156525", "inactivity": "0" }, { "effective_balance": "17000000000", "head": "112851", "target": "112851", "source": "112851", "inclusion_delay": "166307", "inactivity": "0" }, { "effective_balance": "18000000000", "head": "119489", "target": "119489", "source": "119489", "inclusion_delay": "176091", "inactivity": "0" }, { "effective_balance": "19000000000", "head": "126127", "target": "126127", "source": "126127", "inclusion_delay": "185873", "inactivity": "0" }, { "effective_balance": "20000000000", "head": "132766", "target": "132766", "source": "132766", "inclusion_delay": "195656", "inactivity": "0" }, { "effective_balance": "21000000000", "head": "139404", "target": "139404", "source": "139404", "inclusion_delay": "205439", "inactivity": "0" }, { "effective_balance": "22000000000", "head": "146042", "target": "146042", "source": "146042", "inclusion_delay": "215222", "inactivity": "0" }, { "effective_balance": "23000000000", "head": "152681", "target": "152681", "source": "152681", "inclusion_delay": "225004", "inactivity": "0" }, { "effective_balance": "24000000000", "head": "159319", "target": "159319", "source": "159319", "inclusion_delay": "234787", "inactivity": "0" }, { "effective_balance": "25000000000", "head": "165957", "target": "165957", "source": "165957", "inclusion_delay": "244570", "inactivity": "0" }, { "effective_balance": "26000000000", "head": "172596", "target": "172596", "source": "172596", "inclusion_delay": "254352", "inactivity": "0" }, { "effective_balance": "27000000000", "head": "179234", "target": "179234", "source": "179234", "inclusion_delay": "264136", "inactivity": "0" }, { "effective_balance": "28000000000", "head": "185872", "target": "185872", "source": "185872", "inclusion_delay": "273918", "inactivity": "0" }, { "effective_balance": "29000000000", "head": "192510", "target": "192510", "source": "192510", "inclusion_delay": "283701", "inactivity": "0" }, { "effective_balance": "30000000000", "head": "199149", "target": "199149", "source": "199149", "inclusion_delay": "293484", "inactivity": "0" }, { "effective_balance": "31000000000", "head": "205787", "target": "205787", "source": "205787", "inclusion_delay": "303267", "inactivity": "0" }, { "effective_balance": "32000000000", "head": "212426", "target": "212426", "source": "212426", "inclusion_delay": "313050", "inactivity": "0" } ], "total_rewards": [ { "validator_index": "0", "head": "212426", "target": "212426", "source": "212426", "inclusion_delay": "313050", "inactivity": "0" }, { "validator_index": "32", "head": "212426", "target": "212426", "source": "212426", "inclusion_delay": "313050", "inactivity": "0" }, { "validator_index": "63", "head": "-357771", "target": "-357771", "source": "-357771", "inclusion_delay": "0", "inactivity": "0" } ] } ``` </details>
2023-07-18 01:48:40 +00:00
use types::beacon_state::Error as BeaconStateError;
use types::{BeaconState, ChainSpec, ForkName, Slot};
pub const VALIDATOR_COUNT: usize = 64;
Phase 0 attestation rewards via Beacon API (#4474) ## Issue Addressed Addresses #4026. Beacon-API spec [here](https://ethereum.github.io/beacon-APIs/?urls.primaryName=dev#/Beacon/getAttestationsRewards). Endpoint: `POST /eth/v1/beacon/rewards/attestations/{epoch}` This endpoint already supports post-Altair epochs. This PR adds support for phase 0 rewards calculation. ## Proposed Changes - [x] Attestation rewards API to support phase 0 rewards calculation, re-using logic from `state_processing`. Refactored `get_attestation_deltas` slightly to support computing deltas for a subset of validators. - [x] Add `inclusion_delay` to `ideal_rewards` (`beacon-API` spec update to follow) - [x] Add `inactivity` penalties to both `ideal_rewards` and `total_rewards` (`beacon-API` spec update to follow) - [x] Add tests to compute attestation rewards and compare results with beacon states ## Additional Notes - The extra penalty for missing attestations or being slashed during an inactivity leak is currently not included in the API response (for both phase 0 and Altair) in the spec. - I went with adding `inactivity` as a separate component rather than combining them with the 4 rewards, because this is how it was grouped in [the phase 0 spec](https://github.com/ethereum/consensus-specs/blob/dev/specs/phase0/beacon-chain.md#get_attestation_deltas). During inactivity leak, all rewards include the optimal reward, and inactivity penalties are calculated separately (see below code snippet from the spec), so it would be quite confusing if we merge them. This would also work better with Altair, because there's no "cancelling" of rewards and inactivity penalties are more separate. - Altair calculation logic (to include inactivity penalties) to be updated in a follow-up PR. ```python def get_attestation_deltas(state: BeaconState) -> Tuple[Sequence[Gwei], Sequence[Gwei]]: """ Return attestation reward/penalty deltas for each validator. """ source_rewards, source_penalties = get_source_deltas(state) target_rewards, target_penalties = get_target_deltas(state) head_rewards, head_penalties = get_head_deltas(state) inclusion_delay_rewards, _ = get_inclusion_delay_deltas(state) _, inactivity_penalties = get_inactivity_penalty_deltas(state) rewards = [ source_rewards[i] + target_rewards[i] + head_rewards[i] + inclusion_delay_rewards[i] for i in range(len(state.validators)) ] penalties = [ source_penalties[i] + target_penalties[i] + head_penalties[i] + inactivity_penalties[i] for i in range(len(state.validators)) ] return rewards, penalties ``` ## Example API Response <details> <summary>Click me</summary> ```json { "ideal_rewards": [ { "effective_balance": "1000000000", "head": "6638", "target": "6638", "source": "6638", "inclusion_delay": "9783", "inactivity": "0" }, { "effective_balance": "2000000000", "head": "13276", "target": "13276", "source": "13276", "inclusion_delay": "19565", "inactivity": "0" }, { "effective_balance": "3000000000", "head": "19914", "target": "19914", "source": "19914", "inclusion_delay": "29349", "inactivity": "0" }, { "effective_balance": "4000000000", "head": "26553", "target": "26553", "source": "26553", "inclusion_delay": "39131", "inactivity": "0" }, { "effective_balance": "5000000000", "head": "33191", "target": "33191", "source": "33191", "inclusion_delay": "48914", "inactivity": "0" }, { "effective_balance": "6000000000", "head": "39829", "target": "39829", "source": "39829", "inclusion_delay": "58697", "inactivity": "0" }, { "effective_balance": "7000000000", "head": "46468", "target": "46468", "source": "46468", "inclusion_delay": "68480", "inactivity": "0" }, { "effective_balance": "8000000000", "head": "53106", "target": "53106", "source": "53106", "inclusion_delay": "78262", "inactivity": "0" }, { "effective_balance": "9000000000", "head": "59744", "target": "59744", "source": "59744", "inclusion_delay": "88046", "inactivity": "0" }, { "effective_balance": "10000000000", "head": "66383", "target": "66383", "source": "66383", "inclusion_delay": "97828", "inactivity": "0" }, { "effective_balance": "11000000000", "head": "73021", "target": "73021", "source": "73021", "inclusion_delay": "107611", "inactivity": "0" }, { "effective_balance": "12000000000", "head": "79659", "target": "79659", "source": "79659", "inclusion_delay": "117394", "inactivity": "0" }, { "effective_balance": "13000000000", "head": "86298", "target": "86298", "source": "86298", "inclusion_delay": "127176", "inactivity": "0" }, { "effective_balance": "14000000000", "head": "92936", "target": "92936", "source": "92936", "inclusion_delay": "136959", "inactivity": "0" }, { "effective_balance": "15000000000", "head": "99574", "target": "99574", "source": "99574", "inclusion_delay": "146742", "inactivity": "0" }, { "effective_balance": "16000000000", "head": "106212", "target": "106212", "source": "106212", "inclusion_delay": "156525", "inactivity": "0" }, { "effective_balance": "17000000000", "head": "112851", "target": "112851", "source": "112851", "inclusion_delay": "166307", "inactivity": "0" }, { "effective_balance": "18000000000", "head": "119489", "target": "119489", "source": "119489", "inclusion_delay": "176091", "inactivity": "0" }, { "effective_balance": "19000000000", "head": "126127", "target": "126127", "source": "126127", "inclusion_delay": "185873", "inactivity": "0" }, { "effective_balance": "20000000000", "head": "132766", "target": "132766", "source": "132766", "inclusion_delay": "195656", "inactivity": "0" }, { "effective_balance": "21000000000", "head": "139404", "target": "139404", "source": "139404", "inclusion_delay": "205439", "inactivity": "0" }, { "effective_balance": "22000000000", "head": "146042", "target": "146042", "source": "146042", "inclusion_delay": "215222", "inactivity": "0" }, { "effective_balance": "23000000000", "head": "152681", "target": "152681", "source": "152681", "inclusion_delay": "225004", "inactivity": "0" }, { "effective_balance": "24000000000", "head": "159319", "target": "159319", "source": "159319", "inclusion_delay": "234787", "inactivity": "0" }, { "effective_balance": "25000000000", "head": "165957", "target": "165957", "source": "165957", "inclusion_delay": "244570", "inactivity": "0" }, { "effective_balance": "26000000000", "head": "172596", "target": "172596", "source": "172596", "inclusion_delay": "254352", "inactivity": "0" }, { "effective_balance": "27000000000", "head": "179234", "target": "179234", "source": "179234", "inclusion_delay": "264136", "inactivity": "0" }, { "effective_balance": "28000000000", "head": "185872", "target": "185872", "source": "185872", "inclusion_delay": "273918", "inactivity": "0" }, { "effective_balance": "29000000000", "head": "192510", "target": "192510", "source": "192510", "inclusion_delay": "283701", "inactivity": "0" }, { "effective_balance": "30000000000", "head": "199149", "target": "199149", "source": "199149", "inclusion_delay": "293484", "inactivity": "0" }, { "effective_balance": "31000000000", "head": "205787", "target": "205787", "source": "205787", "inclusion_delay": "303267", "inactivity": "0" }, { "effective_balance": "32000000000", "head": "212426", "target": "212426", "source": "212426", "inclusion_delay": "313050", "inactivity": "0" } ], "total_rewards": [ { "validator_index": "0", "head": "212426", "target": "212426", "source": "212426", "inclusion_delay": "313050", "inactivity": "0" }, { "validator_index": "32", "head": "212426", "target": "212426", "source": "212426", "inclusion_delay": "313050", "inactivity": "0" }, { "validator_index": "63", "head": "-357771", "target": "-357771", "source": "-357771", "inclusion_delay": "0", "inactivity": "0" } ] } ``` </details>
2023-07-18 01:48:40 +00:00
type E = MinimalEthSpec;
lazy_static! {
static ref KEYPAIRS: Vec<Keypair> = generate_deterministic_keypairs(VALIDATOR_COUNT);
}
Phase 0 attestation rewards via Beacon API (#4474) ## Issue Addressed Addresses #4026. Beacon-API spec [here](https://ethereum.github.io/beacon-APIs/?urls.primaryName=dev#/Beacon/getAttestationsRewards). Endpoint: `POST /eth/v1/beacon/rewards/attestations/{epoch}` This endpoint already supports post-Altair epochs. This PR adds support for phase 0 rewards calculation. ## Proposed Changes - [x] Attestation rewards API to support phase 0 rewards calculation, re-using logic from `state_processing`. Refactored `get_attestation_deltas` slightly to support computing deltas for a subset of validators. - [x] Add `inclusion_delay` to `ideal_rewards` (`beacon-API` spec update to follow) - [x] Add `inactivity` penalties to both `ideal_rewards` and `total_rewards` (`beacon-API` spec update to follow) - [x] Add tests to compute attestation rewards and compare results with beacon states ## Additional Notes - The extra penalty for missing attestations or being slashed during an inactivity leak is currently not included in the API response (for both phase 0 and Altair) in the spec. - I went with adding `inactivity` as a separate component rather than combining them with the 4 rewards, because this is how it was grouped in [the phase 0 spec](https://github.com/ethereum/consensus-specs/blob/dev/specs/phase0/beacon-chain.md#get_attestation_deltas). During inactivity leak, all rewards include the optimal reward, and inactivity penalties are calculated separately (see below code snippet from the spec), so it would be quite confusing if we merge them. This would also work better with Altair, because there's no "cancelling" of rewards and inactivity penalties are more separate. - Altair calculation logic (to include inactivity penalties) to be updated in a follow-up PR. ```python def get_attestation_deltas(state: BeaconState) -> Tuple[Sequence[Gwei], Sequence[Gwei]]: """ Return attestation reward/penalty deltas for each validator. """ source_rewards, source_penalties = get_source_deltas(state) target_rewards, target_penalties = get_target_deltas(state) head_rewards, head_penalties = get_head_deltas(state) inclusion_delay_rewards, _ = get_inclusion_delay_deltas(state) _, inactivity_penalties = get_inactivity_penalty_deltas(state) rewards = [ source_rewards[i] + target_rewards[i] + head_rewards[i] + inclusion_delay_rewards[i] for i in range(len(state.validators)) ] penalties = [ source_penalties[i] + target_penalties[i] + head_penalties[i] + inactivity_penalties[i] for i in range(len(state.validators)) ] return rewards, penalties ``` ## Example API Response <details> <summary>Click me</summary> ```json { "ideal_rewards": [ { "effective_balance": "1000000000", "head": "6638", "target": "6638", "source": "6638", "inclusion_delay": "9783", "inactivity": "0" }, { "effective_balance": "2000000000", "head": "13276", "target": "13276", "source": "13276", "inclusion_delay": "19565", "inactivity": "0" }, { "effective_balance": "3000000000", "head": "19914", "target": "19914", "source": "19914", "inclusion_delay": "29349", "inactivity": "0" }, { "effective_balance": "4000000000", "head": "26553", "target": "26553", "source": "26553", "inclusion_delay": "39131", "inactivity": "0" }, { "effective_balance": "5000000000", "head": "33191", "target": "33191", "source": "33191", "inclusion_delay": "48914", "inactivity": "0" }, { "effective_balance": "6000000000", "head": "39829", "target": "39829", "source": "39829", "inclusion_delay": "58697", "inactivity": "0" }, { "effective_balance": "7000000000", "head": "46468", "target": "46468", "source": "46468", "inclusion_delay": "68480", "inactivity": "0" }, { "effective_balance": "8000000000", "head": "53106", "target": "53106", "source": "53106", "inclusion_delay": "78262", "inactivity": "0" }, { "effective_balance": "9000000000", "head": "59744", "target": "59744", "source": "59744", "inclusion_delay": "88046", "inactivity": "0" }, { "effective_balance": "10000000000", "head": "66383", "target": "66383", "source": "66383", "inclusion_delay": "97828", "inactivity": "0" }, { "effective_balance": "11000000000", "head": "73021", "target": "73021", "source": "73021", "inclusion_delay": "107611", "inactivity": "0" }, { "effective_balance": "12000000000", "head": "79659", "target": "79659", "source": "79659", "inclusion_delay": "117394", "inactivity": "0" }, { "effective_balance": "13000000000", "head": "86298", "target": "86298", "source": "86298", "inclusion_delay": "127176", "inactivity": "0" }, { "effective_balance": "14000000000", "head": "92936", "target": "92936", "source": "92936", "inclusion_delay": "136959", "inactivity": "0" }, { "effective_balance": "15000000000", "head": "99574", "target": "99574", "source": "99574", "inclusion_delay": "146742", "inactivity": "0" }, { "effective_balance": "16000000000", "head": "106212", "target": "106212", "source": "106212", "inclusion_delay": "156525", "inactivity": "0" }, { "effective_balance": "17000000000", "head": "112851", "target": "112851", "source": "112851", "inclusion_delay": "166307", "inactivity": "0" }, { "effective_balance": "18000000000", "head": "119489", "target": "119489", "source": "119489", "inclusion_delay": "176091", "inactivity": "0" }, { "effective_balance": "19000000000", "head": "126127", "target": "126127", "source": "126127", "inclusion_delay": "185873", "inactivity": "0" }, { "effective_balance": "20000000000", "head": "132766", "target": "132766", "source": "132766", "inclusion_delay": "195656", "inactivity": "0" }, { "effective_balance": "21000000000", "head": "139404", "target": "139404", "source": "139404", "inclusion_delay": "205439", "inactivity": "0" }, { "effective_balance": "22000000000", "head": "146042", "target": "146042", "source": "146042", "inclusion_delay": "215222", "inactivity": "0" }, { "effective_balance": "23000000000", "head": "152681", "target": "152681", "source": "152681", "inclusion_delay": "225004", "inactivity": "0" }, { "effective_balance": "24000000000", "head": "159319", "target": "159319", "source": "159319", "inclusion_delay": "234787", "inactivity": "0" }, { "effective_balance": "25000000000", "head": "165957", "target": "165957", "source": "165957", "inclusion_delay": "244570", "inactivity": "0" }, { "effective_balance": "26000000000", "head": "172596", "target": "172596", "source": "172596", "inclusion_delay": "254352", "inactivity": "0" }, { "effective_balance": "27000000000", "head": "179234", "target": "179234", "source": "179234", "inclusion_delay": "264136", "inactivity": "0" }, { "effective_balance": "28000000000", "head": "185872", "target": "185872", "source": "185872", "inclusion_delay": "273918", "inactivity": "0" }, { "effective_balance": "29000000000", "head": "192510", "target": "192510", "source": "192510", "inclusion_delay": "283701", "inactivity": "0" }, { "effective_balance": "30000000000", "head": "199149", "target": "199149", "source": "199149", "inclusion_delay": "293484", "inactivity": "0" }, { "effective_balance": "31000000000", "head": "205787", "target": "205787", "source": "205787", "inclusion_delay": "303267", "inactivity": "0" }, { "effective_balance": "32000000000", "head": "212426", "target": "212426", "source": "212426", "inclusion_delay": "313050", "inactivity": "0" } ], "total_rewards": [ { "validator_index": "0", "head": "212426", "target": "212426", "source": "212426", "inclusion_delay": "313050", "inactivity": "0" }, { "validator_index": "32", "head": "212426", "target": "212426", "source": "212426", "inclusion_delay": "313050", "inactivity": "0" }, { "validator_index": "63", "head": "-357771", "target": "-357771", "source": "-357771", "inclusion_delay": "0", "inactivity": "0" } ] } ``` </details>
2023-07-18 01:48:40 +00:00
fn get_harness(spec: ChainSpec) -> BeaconChainHarness<EphemeralHarnessType<E>> {
let harness = BeaconChainHarness::builder(E::default())
.spec(spec)
.keypairs(KEYPAIRS.to_vec())
.fresh_ephemeral_store()
.build();
harness.advance_slot();
harness
}
#[tokio::test]
async fn test_sync_committee_rewards() {
Phase 0 attestation rewards via Beacon API (#4474) ## Issue Addressed Addresses #4026. Beacon-API spec [here](https://ethereum.github.io/beacon-APIs/?urls.primaryName=dev#/Beacon/getAttestationsRewards). Endpoint: `POST /eth/v1/beacon/rewards/attestations/{epoch}` This endpoint already supports post-Altair epochs. This PR adds support for phase 0 rewards calculation. ## Proposed Changes - [x] Attestation rewards API to support phase 0 rewards calculation, re-using logic from `state_processing`. Refactored `get_attestation_deltas` slightly to support computing deltas for a subset of validators. - [x] Add `inclusion_delay` to `ideal_rewards` (`beacon-API` spec update to follow) - [x] Add `inactivity` penalties to both `ideal_rewards` and `total_rewards` (`beacon-API` spec update to follow) - [x] Add tests to compute attestation rewards and compare results with beacon states ## Additional Notes - The extra penalty for missing attestations or being slashed during an inactivity leak is currently not included in the API response (for both phase 0 and Altair) in the spec. - I went with adding `inactivity` as a separate component rather than combining them with the 4 rewards, because this is how it was grouped in [the phase 0 spec](https://github.com/ethereum/consensus-specs/blob/dev/specs/phase0/beacon-chain.md#get_attestation_deltas). During inactivity leak, all rewards include the optimal reward, and inactivity penalties are calculated separately (see below code snippet from the spec), so it would be quite confusing if we merge them. This would also work better with Altair, because there's no "cancelling" of rewards and inactivity penalties are more separate. - Altair calculation logic (to include inactivity penalties) to be updated in a follow-up PR. ```python def get_attestation_deltas(state: BeaconState) -> Tuple[Sequence[Gwei], Sequence[Gwei]]: """ Return attestation reward/penalty deltas for each validator. """ source_rewards, source_penalties = get_source_deltas(state) target_rewards, target_penalties = get_target_deltas(state) head_rewards, head_penalties = get_head_deltas(state) inclusion_delay_rewards, _ = get_inclusion_delay_deltas(state) _, inactivity_penalties = get_inactivity_penalty_deltas(state) rewards = [ source_rewards[i] + target_rewards[i] + head_rewards[i] + inclusion_delay_rewards[i] for i in range(len(state.validators)) ] penalties = [ source_penalties[i] + target_penalties[i] + head_penalties[i] + inactivity_penalties[i] for i in range(len(state.validators)) ] return rewards, penalties ``` ## Example API Response <details> <summary>Click me</summary> ```json { "ideal_rewards": [ { "effective_balance": "1000000000", "head": "6638", "target": "6638", "source": "6638", "inclusion_delay": "9783", "inactivity": "0" }, { "effective_balance": "2000000000", "head": "13276", "target": "13276", "source": "13276", "inclusion_delay": "19565", "inactivity": "0" }, { "effective_balance": "3000000000", "head": "19914", "target": "19914", "source": "19914", "inclusion_delay": "29349", "inactivity": "0" }, { "effective_balance": "4000000000", "head": "26553", "target": "26553", "source": "26553", "inclusion_delay": "39131", "inactivity": "0" }, { "effective_balance": "5000000000", "head": "33191", "target": "33191", "source": "33191", "inclusion_delay": "48914", "inactivity": "0" }, { "effective_balance": "6000000000", "head": "39829", "target": "39829", "source": "39829", "inclusion_delay": "58697", "inactivity": "0" }, { "effective_balance": "7000000000", "head": "46468", "target": "46468", "source": "46468", "inclusion_delay": "68480", "inactivity": "0" }, { "effective_balance": "8000000000", "head": "53106", "target": "53106", "source": "53106", "inclusion_delay": "78262", "inactivity": "0" }, { "effective_balance": "9000000000", "head": "59744", "target": "59744", "source": "59744", "inclusion_delay": "88046", "inactivity": "0" }, { "effective_balance": "10000000000", "head": "66383", "target": "66383", "source": "66383", "inclusion_delay": "97828", "inactivity": "0" }, { "effective_balance": "11000000000", "head": "73021", "target": "73021", "source": "73021", "inclusion_delay": "107611", "inactivity": "0" }, { "effective_balance": "12000000000", "head": "79659", "target": "79659", "source": "79659", "inclusion_delay": "117394", "inactivity": "0" }, { "effective_balance": "13000000000", "head": "86298", "target": "86298", "source": "86298", "inclusion_delay": "127176", "inactivity": "0" }, { "effective_balance": "14000000000", "head": "92936", "target": "92936", "source": "92936", "inclusion_delay": "136959", "inactivity": "0" }, { "effective_balance": "15000000000", "head": "99574", "target": "99574", "source": "99574", "inclusion_delay": "146742", "inactivity": "0" }, { "effective_balance": "16000000000", "head": "106212", "target": "106212", "source": "106212", "inclusion_delay": "156525", "inactivity": "0" }, { "effective_balance": "17000000000", "head": "112851", "target": "112851", "source": "112851", "inclusion_delay": "166307", "inactivity": "0" }, { "effective_balance": "18000000000", "head": "119489", "target": "119489", "source": "119489", "inclusion_delay": "176091", "inactivity": "0" }, { "effective_balance": "19000000000", "head": "126127", "target": "126127", "source": "126127", "inclusion_delay": "185873", "inactivity": "0" }, { "effective_balance": "20000000000", "head": "132766", "target": "132766", "source": "132766", "inclusion_delay": "195656", "inactivity": "0" }, { "effective_balance": "21000000000", "head": "139404", "target": "139404", "source": "139404", "inclusion_delay": "205439", "inactivity": "0" }, { "effective_balance": "22000000000", "head": "146042", "target": "146042", "source": "146042", "inclusion_delay": "215222", "inactivity": "0" }, { "effective_balance": "23000000000", "head": "152681", "target": "152681", "source": "152681", "inclusion_delay": "225004", "inactivity": "0" }, { "effective_balance": "24000000000", "head": "159319", "target": "159319", "source": "159319", "inclusion_delay": "234787", "inactivity": "0" }, { "effective_balance": "25000000000", "head": "165957", "target": "165957", "source": "165957", "inclusion_delay": "244570", "inactivity": "0" }, { "effective_balance": "26000000000", "head": "172596", "target": "172596", "source": "172596", "inclusion_delay": "254352", "inactivity": "0" }, { "effective_balance": "27000000000", "head": "179234", "target": "179234", "source": "179234", "inclusion_delay": "264136", "inactivity": "0" }, { "effective_balance": "28000000000", "head": "185872", "target": "185872", "source": "185872", "inclusion_delay": "273918", "inactivity": "0" }, { "effective_balance": "29000000000", "head": "192510", "target": "192510", "source": "192510", "inclusion_delay": "283701", "inactivity": "0" }, { "effective_balance": "30000000000", "head": "199149", "target": "199149", "source": "199149", "inclusion_delay": "293484", "inactivity": "0" }, { "effective_balance": "31000000000", "head": "205787", "target": "205787", "source": "205787", "inclusion_delay": "303267", "inactivity": "0" }, { "effective_balance": "32000000000", "head": "212426", "target": "212426", "source": "212426", "inclusion_delay": "313050", "inactivity": "0" } ], "total_rewards": [ { "validator_index": "0", "head": "212426", "target": "212426", "source": "212426", "inclusion_delay": "313050", "inactivity": "0" }, { "validator_index": "32", "head": "212426", "target": "212426", "source": "212426", "inclusion_delay": "313050", "inactivity": "0" }, { "validator_index": "63", "head": "-357771", "target": "-357771", "source": "-357771", "inclusion_delay": "0", "inactivity": "0" } ] } ``` </details>
2023-07-18 01:48:40 +00:00
let mut spec = E::default_spec();
spec.altair_fork_epoch = Some(Epoch::new(0));
let harness = get_harness(spec);
let num_block_produced = E::slots_per_epoch();
let latest_block_root = harness
.extend_chain(
num_block_produced as usize,
BlockStrategy::OnCanonicalHead,
AttestationStrategy::AllValidators,
)
.await;
// Create and add sync committee message to op_pool
let sync_contributions = harness.make_sync_contributions(
&harness.get_current_state(),
latest_block_root,
harness.get_current_slot(),
RelativeSyncCommittee::Current,
);
harness
.process_sync_contributions(sync_contributions)
.unwrap();
// Add block
let chain = &harness.chain;
let (head_state, head_state_root) = harness.get_current_state_and_root();
let target_slot = harness.get_current_slot() + 1;
let (block_root, mut state) = harness
.add_attested_block_at_slot(target_slot, head_state, head_state_root, &[])
.await
.unwrap();
let block = harness.get_block(block_root).unwrap();
let parent_block = chain
.get_blinded_block(&block.parent_root())
.unwrap()
.unwrap();
let parent_state = chain
.get_state(&parent_block.state_root(), Some(parent_block.slot()))
.unwrap()
.unwrap();
let reward_payload = chain
.compute_sync_committee_rewards(block.message(), &mut state)
.unwrap();
let rewards = reward_payload
.iter()
.map(|reward| (reward.validator_index, reward.reward))
.collect::<HashMap<_, _>>();
let proposer_index = state
.get_beacon_proposer_index(target_slot, &MinimalEthSpec::default_spec())
.unwrap();
let mut mismatches = vec![];
for validator in state.validators() {
let validator_index = state
.clone()
.get_validator_index(&validator.pubkey)
.unwrap()
.unwrap();
let pre_state_balance = parent_state.balances()[validator_index];
let post_state_balance = state.balances()[validator_index];
let sync_committee_reward = rewards.get(&(validator_index as u64)).unwrap_or(&0);
if validator_index == proposer_index {
continue; // Ignore proposer
}
if pre_state_balance as i64 + *sync_committee_reward != post_state_balance as i64 {
mismatches.push(validator_index.to_string());
}
}
assert_eq!(
mismatches.len(),
0,
"Expect 0 mismatches, but these validators have mismatches on balance: {} ",
mismatches.join(",")
);
}
Phase 0 attestation rewards via Beacon API (#4474) ## Issue Addressed Addresses #4026. Beacon-API spec [here](https://ethereum.github.io/beacon-APIs/?urls.primaryName=dev#/Beacon/getAttestationsRewards). Endpoint: `POST /eth/v1/beacon/rewards/attestations/{epoch}` This endpoint already supports post-Altair epochs. This PR adds support for phase 0 rewards calculation. ## Proposed Changes - [x] Attestation rewards API to support phase 0 rewards calculation, re-using logic from `state_processing`. Refactored `get_attestation_deltas` slightly to support computing deltas for a subset of validators. - [x] Add `inclusion_delay` to `ideal_rewards` (`beacon-API` spec update to follow) - [x] Add `inactivity` penalties to both `ideal_rewards` and `total_rewards` (`beacon-API` spec update to follow) - [x] Add tests to compute attestation rewards and compare results with beacon states ## Additional Notes - The extra penalty for missing attestations or being slashed during an inactivity leak is currently not included in the API response (for both phase 0 and Altair) in the spec. - I went with adding `inactivity` as a separate component rather than combining them with the 4 rewards, because this is how it was grouped in [the phase 0 spec](https://github.com/ethereum/consensus-specs/blob/dev/specs/phase0/beacon-chain.md#get_attestation_deltas). During inactivity leak, all rewards include the optimal reward, and inactivity penalties are calculated separately (see below code snippet from the spec), so it would be quite confusing if we merge them. This would also work better with Altair, because there's no "cancelling" of rewards and inactivity penalties are more separate. - Altair calculation logic (to include inactivity penalties) to be updated in a follow-up PR. ```python def get_attestation_deltas(state: BeaconState) -> Tuple[Sequence[Gwei], Sequence[Gwei]]: """ Return attestation reward/penalty deltas for each validator. """ source_rewards, source_penalties = get_source_deltas(state) target_rewards, target_penalties = get_target_deltas(state) head_rewards, head_penalties = get_head_deltas(state) inclusion_delay_rewards, _ = get_inclusion_delay_deltas(state) _, inactivity_penalties = get_inactivity_penalty_deltas(state) rewards = [ source_rewards[i] + target_rewards[i] + head_rewards[i] + inclusion_delay_rewards[i] for i in range(len(state.validators)) ] penalties = [ source_penalties[i] + target_penalties[i] + head_penalties[i] + inactivity_penalties[i] for i in range(len(state.validators)) ] return rewards, penalties ``` ## Example API Response <details> <summary>Click me</summary> ```json { "ideal_rewards": [ { "effective_balance": "1000000000", "head": "6638", "target": "6638", "source": "6638", "inclusion_delay": "9783", "inactivity": "0" }, { "effective_balance": "2000000000", "head": "13276", "target": "13276", "source": "13276", "inclusion_delay": "19565", "inactivity": "0" }, { "effective_balance": "3000000000", "head": "19914", "target": "19914", "source": "19914", "inclusion_delay": "29349", "inactivity": "0" }, { "effective_balance": "4000000000", "head": "26553", "target": "26553", "source": "26553", "inclusion_delay": "39131", "inactivity": "0" }, { "effective_balance": "5000000000", "head": "33191", "target": "33191", "source": "33191", "inclusion_delay": "48914", "inactivity": "0" }, { "effective_balance": "6000000000", "head": "39829", "target": "39829", "source": "39829", "inclusion_delay": "58697", "inactivity": "0" }, { "effective_balance": "7000000000", "head": "46468", "target": "46468", "source": "46468", "inclusion_delay": "68480", "inactivity": "0" }, { "effective_balance": "8000000000", "head": "53106", "target": "53106", "source": "53106", "inclusion_delay": "78262", "inactivity": "0" }, { "effective_balance": "9000000000", "head": "59744", "target": "59744", "source": "59744", "inclusion_delay": "88046", "inactivity": "0" }, { "effective_balance": "10000000000", "head": "66383", "target": "66383", "source": "66383", "inclusion_delay": "97828", "inactivity": "0" }, { "effective_balance": "11000000000", "head": "73021", "target": "73021", "source": "73021", "inclusion_delay": "107611", "inactivity": "0" }, { "effective_balance": "12000000000", "head": "79659", "target": "79659", "source": "79659", "inclusion_delay": "117394", "inactivity": "0" }, { "effective_balance": "13000000000", "head": "86298", "target": "86298", "source": "86298", "inclusion_delay": "127176", "inactivity": "0" }, { "effective_balance": "14000000000", "head": "92936", "target": "92936", "source": "92936", "inclusion_delay": "136959", "inactivity": "0" }, { "effective_balance": "15000000000", "head": "99574", "target": "99574", "source": "99574", "inclusion_delay": "146742", "inactivity": "0" }, { "effective_balance": "16000000000", "head": "106212", "target": "106212", "source": "106212", "inclusion_delay": "156525", "inactivity": "0" }, { "effective_balance": "17000000000", "head": "112851", "target": "112851", "source": "112851", "inclusion_delay": "166307", "inactivity": "0" }, { "effective_balance": "18000000000", "head": "119489", "target": "119489", "source": "119489", "inclusion_delay": "176091", "inactivity": "0" }, { "effective_balance": "19000000000", "head": "126127", "target": "126127", "source": "126127", "inclusion_delay": "185873", "inactivity": "0" }, { "effective_balance": "20000000000", "head": "132766", "target": "132766", "source": "132766", "inclusion_delay": "195656", "inactivity": "0" }, { "effective_balance": "21000000000", "head": "139404", "target": "139404", "source": "139404", "inclusion_delay": "205439", "inactivity": "0" }, { "effective_balance": "22000000000", "head": "146042", "target": "146042", "source": "146042", "inclusion_delay": "215222", "inactivity": "0" }, { "effective_balance": "23000000000", "head": "152681", "target": "152681", "source": "152681", "inclusion_delay": "225004", "inactivity": "0" }, { "effective_balance": "24000000000", "head": "159319", "target": "159319", "source": "159319", "inclusion_delay": "234787", "inactivity": "0" }, { "effective_balance": "25000000000", "head": "165957", "target": "165957", "source": "165957", "inclusion_delay": "244570", "inactivity": "0" }, { "effective_balance": "26000000000", "head": "172596", "target": "172596", "source": "172596", "inclusion_delay": "254352", "inactivity": "0" }, { "effective_balance": "27000000000", "head": "179234", "target": "179234", "source": "179234", "inclusion_delay": "264136", "inactivity": "0" }, { "effective_balance": "28000000000", "head": "185872", "target": "185872", "source": "185872", "inclusion_delay": "273918", "inactivity": "0" }, { "effective_balance": "29000000000", "head": "192510", "target": "192510", "source": "192510", "inclusion_delay": "283701", "inactivity": "0" }, { "effective_balance": "30000000000", "head": "199149", "target": "199149", "source": "199149", "inclusion_delay": "293484", "inactivity": "0" }, { "effective_balance": "31000000000", "head": "205787", "target": "205787", "source": "205787", "inclusion_delay": "303267", "inactivity": "0" }, { "effective_balance": "32000000000", "head": "212426", "target": "212426", "source": "212426", "inclusion_delay": "313050", "inactivity": "0" } ], "total_rewards": [ { "validator_index": "0", "head": "212426", "target": "212426", "source": "212426", "inclusion_delay": "313050", "inactivity": "0" }, { "validator_index": "32", "head": "212426", "target": "212426", "source": "212426", "inclusion_delay": "313050", "inactivity": "0" }, { "validator_index": "63", "head": "-357771", "target": "-357771", "source": "-357771", "inclusion_delay": "0", "inactivity": "0" } ] } ``` </details>
2023-07-18 01:48:40 +00:00
#[tokio::test]
async fn test_verify_attestation_rewards_base() {
let harness = get_harness(E::default_spec());
// epoch 0 (N), only two thirds of validators vote.
let two_thirds = (VALIDATOR_COUNT / 3) * 2;
let two_thirds_validators: Vec<usize> = (0..two_thirds).collect();
harness
.extend_chain(
E::slots_per_epoch() as usize,
BlockStrategy::OnCanonicalHead,
AttestationStrategy::SomeValidators(two_thirds_validators),
)
.await;
let initial_balances: Vec<u64> = harness.get_current_state().balances().clone().into();
// extend slots to beginning of epoch N + 2
harness.extend_slots(E::slots_per_epoch() as usize).await;
// compute reward deltas for all validators in epoch N
let StandardAttestationRewards {
ideal_rewards,
total_rewards,
} = harness
.chain
.compute_attestation_rewards(Epoch::new(0), vec![])
.unwrap();
// assert no inactivity penalty for both ideal rewards and individual validators
assert!(ideal_rewards.iter().all(|reward| reward.inactivity == 0));
assert!(total_rewards.iter().all(|reward| reward.inactivity == 0));
// apply attestation rewards to initial balances
let expected_balances = apply_attestation_rewards(&initial_balances, total_rewards);
// verify expected balances against actual balances
let balances: Vec<u64> = harness.get_current_state().balances().clone().into();
assert_eq!(expected_balances, balances);
}
#[tokio::test]
async fn test_verify_attestation_rewards_base_inactivity_leak() {
let spec = E::default_spec();
let harness = get_harness(spec.clone());
let half = VALIDATOR_COUNT / 2;
let half_validators: Vec<usize> = (0..half).collect();
// target epoch is the epoch where the chain enters inactivity leak
let target_epoch = &spec.min_epochs_to_inactivity_penalty + 1;
// advance until beginning of epoch N + 1 and get balances
harness
.extend_chain(
(E::slots_per_epoch() * (target_epoch + 1)) as usize,
BlockStrategy::OnCanonicalHead,
AttestationStrategy::SomeValidators(half_validators.clone()),
)
.await;
let initial_balances: Vec<u64> = harness.get_current_state().balances().clone().into();
// extend slots to beginning of epoch N + 2
harness.advance_slot();
harness
.extend_chain(
E::slots_per_epoch() as usize,
BlockStrategy::OnCanonicalHead,
AttestationStrategy::SomeValidators(half_validators),
)
.await;
let _slot = harness.get_current_slot();
// compute reward deltas for all validators in epoch N
let StandardAttestationRewards {
ideal_rewards,
total_rewards,
} = harness
.chain
.compute_attestation_rewards(Epoch::new(target_epoch), vec![])
.unwrap();
// assert inactivity penalty for both ideal rewards and individual validators
assert!(ideal_rewards.iter().all(|reward| reward.inactivity < 0));
assert!(total_rewards.iter().all(|reward| reward.inactivity < 0));
// apply attestation rewards to initial balances
let expected_balances = apply_attestation_rewards(&initial_balances, total_rewards);
// verify expected balances against actual balances
let balances: Vec<u64> = harness.get_current_state().balances().clone().into();
assert_eq!(expected_balances, balances);
}
#[tokio::test]
async fn test_verify_attestation_rewards_base_inactivity_leak_justification_epoch() {
let spec = E::default_spec();
let harness = get_harness(spec.clone());
let half = VALIDATOR_COUNT / 2;
let half_validators: Vec<usize> = (0..half).collect();
// target epoch is the epoch where the chain enters inactivity leak
let mut target_epoch = &spec.min_epochs_to_inactivity_penalty + 2;
// advance until beginning of epoch N + 2
harness
.extend_chain(
(E::slots_per_epoch() * (target_epoch + 1)) as usize,
BlockStrategy::OnCanonicalHead,
AttestationStrategy::SomeValidators(half_validators.clone()),
)
.await;
// advance to create first justification epoch and get initial balances
harness.extend_slots(E::slots_per_epoch() as usize).await;
target_epoch += 1;
let initial_balances: Vec<u64> = harness.get_current_state().balances().clone().into();
//assert previous_justified_checkpoint matches 0 as we were in inactivity leak from beginning
assert_eq!(
0,
harness
.get_current_state()
.previous_justified_checkpoint()
.epoch
.as_u64()
);
// extend slots to beginning of epoch N + 1
harness.extend_slots(E::slots_per_epoch() as usize).await;
//assert target epoch and previous_justified_checkpoint match
assert_eq!(
target_epoch,
harness
.get_current_state()
.previous_justified_checkpoint()
.epoch
.as_u64()
);
// compute reward deltas for all validators in epoch N
let StandardAttestationRewards {
ideal_rewards,
total_rewards,
} = harness
.chain
.compute_attestation_rewards(Epoch::new(target_epoch), vec![])
.unwrap();
// assert we successfully get ideal rewards for justified epoch out of inactivity leak
assert!(ideal_rewards
.iter()
.all(|reward| reward.head > 0 && reward.target > 0 && reward.source > 0));
// apply attestation rewards to initial balances
let expected_balances = apply_attestation_rewards(&initial_balances, total_rewards);
// verify expected balances against actual balances
let balances: Vec<u64> = harness.get_current_state().balances().clone().into();
assert_eq!(expected_balances, balances);
}
#[tokio::test]
async fn test_verify_attestation_rewards_altair() {
let spec = ForkName::Altair.make_genesis_spec(E::default_spec());
let harness = get_harness(spec.clone());
let target_epoch = 0;
// advance until epoch N + 1 and get initial balances
harness
.extend_slots((E::slots_per_epoch() * (target_epoch + 1)) as usize)
.await;
let initial_balances: Vec<u64> = harness.get_current_state().balances().clone().into();
// advance until epoch N + 2 and build proposal rewards map
let mut proposal_rewards_map: HashMap<u64, u64> = HashMap::new();
let mut sync_committee_rewards_map: HashMap<u64, i64> = HashMap::new();
for _ in 0..E::slots_per_epoch() {
let state = harness.get_current_state();
let slot = state.slot() + Slot::new(1);
// calculate beacon block rewards / penalties
let ((signed_block, _maybe_blob_sidecars), mut state) =
harness.make_block_return_pre_state(state, slot).await;
let beacon_block_reward = harness
.chain
.compute_beacon_block_reward(
signed_block.message(),
signed_block.canonical_root(),
&mut state,
)
.unwrap();
let total_proposer_reward = proposal_rewards_map
.get(&beacon_block_reward.proposer_index)
.unwrap_or(&0u64)
+ beacon_block_reward.total;
proposal_rewards_map.insert(beacon_block_reward.proposer_index, total_proposer_reward);
// calculate sync committee rewards / penalties
let reward_payload = harness
.chain
.compute_sync_committee_rewards(signed_block.message(), &mut state)
.unwrap();
reward_payload.iter().for_each(|reward| {
let mut amount = *sync_committee_rewards_map
.get(&reward.validator_index)
.unwrap_or(&0);
amount += reward.reward;
sync_committee_rewards_map.insert(reward.validator_index, amount);
});
harness.extend_slots(1).await;
}
// compute reward deltas for all validators in epoch N
let StandardAttestationRewards {
ideal_rewards,
total_rewards,
} = harness
.chain
.compute_attestation_rewards(Epoch::new(target_epoch), vec![])
.unwrap();
// assert ideal rewards are greater than 0
assert!(ideal_rewards
.iter()
.all(|reward| reward.head > 0 && reward.target > 0 && reward.source > 0));
// apply attestation, proposal, and sync committee rewards and penalties to initial balances
let expected_balances = apply_attestation_rewards(&initial_balances, total_rewards);
let expected_balances = apply_beacon_block_rewards(&proposal_rewards_map, expected_balances);
let expected_balances =
apply_sync_committee_rewards(&sync_committee_rewards_map, expected_balances);
// verify expected balances against actual balances
let balances: Vec<u64> = harness.get_current_state().balances().clone().into();
assert_eq!(expected_balances, balances);
}
#[tokio::test]
async fn test_verify_attestation_rewards_altair_inactivity_leak() {
let spec = ForkName::Altair.make_genesis_spec(E::default_spec());
let harness = get_harness(spec.clone());
let half = VALIDATOR_COUNT / 2;
let half_validators: Vec<usize> = (0..half).collect();
// target epoch is the epoch where the chain enters inactivity leak
let target_epoch = &spec.min_epochs_to_inactivity_penalty + 1;
// advance until beginning of epoch N + 1 and get balances
harness
.extend_slots_some_validators(
(E::slots_per_epoch() * (target_epoch + 1)) as usize,
half_validators.clone(),
)
.await;
let initial_balances: Vec<u64> = harness.get_current_state().balances().clone().into();
// advance until epoch N + 2 and build proposal rewards map
let mut proposal_rewards_map: HashMap<u64, u64> = HashMap::new();
let mut sync_committee_rewards_map: HashMap<u64, i64> = HashMap::new();
for _ in 0..E::slots_per_epoch() {
let state = harness.get_current_state();
let slot = state.slot() + Slot::new(1);
// calculate beacon block rewards / penalties
let ((signed_block, _maybe_blob_sidecars), mut state) =
harness.make_block_return_pre_state(state, slot).await;
let beacon_block_reward = harness
.chain
.compute_beacon_block_reward(
signed_block.message(),
signed_block.canonical_root(),
&mut state,
)
.unwrap();
let total_proposer_reward = proposal_rewards_map
.get(&beacon_block_reward.proposer_index)
.unwrap_or(&0u64)
+ beacon_block_reward.total;
proposal_rewards_map.insert(beacon_block_reward.proposer_index, total_proposer_reward);
// calculate sync committee rewards / penalties
let reward_payload = harness
.chain
.compute_sync_committee_rewards(signed_block.message(), &mut state)
.unwrap();
reward_payload.iter().for_each(|reward| {
let mut amount = *sync_committee_rewards_map
.get(&reward.validator_index)
.unwrap_or(&0);
amount += reward.reward;
sync_committee_rewards_map.insert(reward.validator_index, amount);
});
harness
.extend_slots_some_validators(1, half_validators.clone())
.await;
}
// compute reward deltas for all validators in epoch N
let StandardAttestationRewards {
ideal_rewards,
total_rewards,
} = harness
.chain
.compute_attestation_rewards(Epoch::new(target_epoch), vec![])
.unwrap();
// assert inactivity penalty for both ideal rewards and individual validators
assert!(ideal_rewards.iter().all(|reward| reward.inactivity == 0));
assert!(total_rewards[..half]
.iter()
.all(|reward| reward.inactivity == 0));
assert!(total_rewards[half..]
.iter()
.all(|reward| reward.inactivity < 0));
// apply attestation, proposal, and sync committee rewards and penalties to initial balances
let expected_balances = apply_attestation_rewards(&initial_balances, total_rewards);
let expected_balances = apply_beacon_block_rewards(&proposal_rewards_map, expected_balances);
let expected_balances =
apply_sync_committee_rewards(&sync_committee_rewards_map, expected_balances);
// verify expected balances against actual balances
let balances: Vec<u64> = harness.get_current_state().balances().clone().into();
assert_eq!(expected_balances, balances);
}
#[tokio::test]
async fn test_verify_attestation_rewards_altair_inactivity_leak_justification_epoch() {
let spec = ForkName::Altair.make_genesis_spec(E::default_spec());
let harness = get_harness(spec.clone());
let half = VALIDATOR_COUNT / 2;
let half_validators: Vec<usize> = (0..half).collect();
// target epoch is the epoch where the chain enters inactivity leak + 1
let mut target_epoch = &spec.min_epochs_to_inactivity_penalty + 2;
// advance until beginning of epoch N + 1
harness
.extend_slots_some_validators(
(E::slots_per_epoch() * (target_epoch + 1)) as usize,
half_validators.clone(),
)
.await;
let validator_inactivity_score = harness
.get_current_state()
.get_inactivity_score(VALIDATOR_COUNT - 1)
.unwrap();
//assert to ensure we are in inactivity leak
assert_eq!(4, validator_inactivity_score);
// advance for first justification epoch and get balances
harness.extend_slots(E::slots_per_epoch() as usize).await;
target_epoch += 1;
let initial_balances: Vec<u64> = harness.get_current_state().balances().clone().into();
// advance until epoch N + 2 and build proposal rewards map
let mut proposal_rewards_map: HashMap<u64, u64> = HashMap::new();
let mut sync_committee_rewards_map: HashMap<u64, i64> = HashMap::new();
for _ in 0..E::slots_per_epoch() {
let state = harness.get_current_state();
let slot = state.slot() + Slot::new(1);
// calculate beacon block rewards / penalties
let ((signed_block, _maybe_blob_sidecars), mut state) =
harness.make_block_return_pre_state(state, slot).await;
let beacon_block_reward = harness
.chain
.compute_beacon_block_reward(
signed_block.message(),
signed_block.canonical_root(),
&mut state,
)
.unwrap();
let total_proposer_reward = proposal_rewards_map
.get(&beacon_block_reward.proposer_index)
.unwrap_or(&0u64)
+ beacon_block_reward.total;
proposal_rewards_map.insert(beacon_block_reward.proposer_index, total_proposer_reward);
// calculate sync committee rewards / penalties
let reward_payload = harness
.chain
.compute_sync_committee_rewards(signed_block.message(), &mut state)
.unwrap();
reward_payload.iter().for_each(|reward| {
let mut amount = *sync_committee_rewards_map
.get(&reward.validator_index)
.unwrap_or(&0);
amount += reward.reward;
sync_committee_rewards_map.insert(reward.validator_index, amount);
});
harness.extend_slots(1).await;
}
//assert target epoch and previous_justified_checkpoint match
assert_eq!(
target_epoch,
harness
.get_current_state()
.previous_justified_checkpoint()
.epoch
.as_u64()
);
// compute reward deltas for all validators in epoch N
let StandardAttestationRewards {
ideal_rewards,
total_rewards,
} = harness
.chain
.compute_attestation_rewards(Epoch::new(target_epoch), vec![])
.unwrap();
// assert ideal rewards are greater than 0
assert!(ideal_rewards
.iter()
.all(|reward| reward.head > 0 && reward.target > 0 && reward.source > 0));
// apply attestation, proposal, and sync committee rewards and penalties to initial balances
let expected_balances = apply_attestation_rewards(&initial_balances, total_rewards);
let expected_balances = apply_beacon_block_rewards(&proposal_rewards_map, expected_balances);
let expected_balances =
apply_sync_committee_rewards(&sync_committee_rewards_map, expected_balances);
// verify expected balances against actual balances
let balances: Vec<u64> = harness.get_current_state().balances().clone().into();
assert_eq!(expected_balances, balances);
}
Phase 0 attestation rewards via Beacon API (#4474) ## Issue Addressed Addresses #4026. Beacon-API spec [here](https://ethereum.github.io/beacon-APIs/?urls.primaryName=dev#/Beacon/getAttestationsRewards). Endpoint: `POST /eth/v1/beacon/rewards/attestations/{epoch}` This endpoint already supports post-Altair epochs. This PR adds support for phase 0 rewards calculation. ## Proposed Changes - [x] Attestation rewards API to support phase 0 rewards calculation, re-using logic from `state_processing`. Refactored `get_attestation_deltas` slightly to support computing deltas for a subset of validators. - [x] Add `inclusion_delay` to `ideal_rewards` (`beacon-API` spec update to follow) - [x] Add `inactivity` penalties to both `ideal_rewards` and `total_rewards` (`beacon-API` spec update to follow) - [x] Add tests to compute attestation rewards and compare results with beacon states ## Additional Notes - The extra penalty for missing attestations or being slashed during an inactivity leak is currently not included in the API response (for both phase 0 and Altair) in the spec. - I went with adding `inactivity` as a separate component rather than combining them with the 4 rewards, because this is how it was grouped in [the phase 0 spec](https://github.com/ethereum/consensus-specs/blob/dev/specs/phase0/beacon-chain.md#get_attestation_deltas). During inactivity leak, all rewards include the optimal reward, and inactivity penalties are calculated separately (see below code snippet from the spec), so it would be quite confusing if we merge them. This would also work better with Altair, because there's no "cancelling" of rewards and inactivity penalties are more separate. - Altair calculation logic (to include inactivity penalties) to be updated in a follow-up PR. ```python def get_attestation_deltas(state: BeaconState) -> Tuple[Sequence[Gwei], Sequence[Gwei]]: """ Return attestation reward/penalty deltas for each validator. """ source_rewards, source_penalties = get_source_deltas(state) target_rewards, target_penalties = get_target_deltas(state) head_rewards, head_penalties = get_head_deltas(state) inclusion_delay_rewards, _ = get_inclusion_delay_deltas(state) _, inactivity_penalties = get_inactivity_penalty_deltas(state) rewards = [ source_rewards[i] + target_rewards[i] + head_rewards[i] + inclusion_delay_rewards[i] for i in range(len(state.validators)) ] penalties = [ source_penalties[i] + target_penalties[i] + head_penalties[i] + inactivity_penalties[i] for i in range(len(state.validators)) ] return rewards, penalties ``` ## Example API Response <details> <summary>Click me</summary> ```json { "ideal_rewards": [ { "effective_balance": "1000000000", "head": "6638", "target": "6638", "source": "6638", "inclusion_delay": "9783", "inactivity": "0" }, { "effective_balance": "2000000000", "head": "13276", "target": "13276", "source": "13276", "inclusion_delay": "19565", "inactivity": "0" }, { "effective_balance": "3000000000", "head": "19914", "target": "19914", "source": "19914", "inclusion_delay": "29349", "inactivity": "0" }, { "effective_balance": "4000000000", "head": "26553", "target": "26553", "source": "26553", "inclusion_delay": "39131", "inactivity": "0" }, { "effective_balance": "5000000000", "head": "33191", "target": "33191", "source": "33191", "inclusion_delay": "48914", "inactivity": "0" }, { "effective_balance": "6000000000", "head": "39829", "target": "39829", "source": "39829", "inclusion_delay": "58697", "inactivity": "0" }, { "effective_balance": "7000000000", "head": "46468", "target": "46468", "source": "46468", "inclusion_delay": "68480", "inactivity": "0" }, { "effective_balance": "8000000000", "head": "53106", "target": "53106", "source": "53106", "inclusion_delay": "78262", "inactivity": "0" }, { "effective_balance": "9000000000", "head": "59744", "target": "59744", "source": "59744", "inclusion_delay": "88046", "inactivity": "0" }, { "effective_balance": "10000000000", "head": "66383", "target": "66383", "source": "66383", "inclusion_delay": "97828", "inactivity": "0" }, { "effective_balance": "11000000000", "head": "73021", "target": "73021", "source": "73021", "inclusion_delay": "107611", "inactivity": "0" }, { "effective_balance": "12000000000", "head": "79659", "target": "79659", "source": "79659", "inclusion_delay": "117394", "inactivity": "0" }, { "effective_balance": "13000000000", "head": "86298", "target": "86298", "source": "86298", "inclusion_delay": "127176", "inactivity": "0" }, { "effective_balance": "14000000000", "head": "92936", "target": "92936", "source": "92936", "inclusion_delay": "136959", "inactivity": "0" }, { "effective_balance": "15000000000", "head": "99574", "target": "99574", "source": "99574", "inclusion_delay": "146742", "inactivity": "0" }, { "effective_balance": "16000000000", "head": "106212", "target": "106212", "source": "106212", "inclusion_delay": "156525", "inactivity": "0" }, { "effective_balance": "17000000000", "head": "112851", "target": "112851", "source": "112851", "inclusion_delay": "166307", "inactivity": "0" }, { "effective_balance": "18000000000", "head": "119489", "target": "119489", "source": "119489", "inclusion_delay": "176091", "inactivity": "0" }, { "effective_balance": "19000000000", "head": "126127", "target": "126127", "source": "126127", "inclusion_delay": "185873", "inactivity": "0" }, { "effective_balance": "20000000000", "head": "132766", "target": "132766", "source": "132766", "inclusion_delay": "195656", "inactivity": "0" }, { "effective_balance": "21000000000", "head": "139404", "target": "139404", "source": "139404", "inclusion_delay": "205439", "inactivity": "0" }, { "effective_balance": "22000000000", "head": "146042", "target": "146042", "source": "146042", "inclusion_delay": "215222", "inactivity": "0" }, { "effective_balance": "23000000000", "head": "152681", "target": "152681", "source": "152681", "inclusion_delay": "225004", "inactivity": "0" }, { "effective_balance": "24000000000", "head": "159319", "target": "159319", "source": "159319", "inclusion_delay": "234787", "inactivity": "0" }, { "effective_balance": "25000000000", "head": "165957", "target": "165957", "source": "165957", "inclusion_delay": "244570", "inactivity": "0" }, { "effective_balance": "26000000000", "head": "172596", "target": "172596", "source": "172596", "inclusion_delay": "254352", "inactivity": "0" }, { "effective_balance": "27000000000", "head": "179234", "target": "179234", "source": "179234", "inclusion_delay": "264136", "inactivity": "0" }, { "effective_balance": "28000000000", "head": "185872", "target": "185872", "source": "185872", "inclusion_delay": "273918", "inactivity": "0" }, { "effective_balance": "29000000000", "head": "192510", "target": "192510", "source": "192510", "inclusion_delay": "283701", "inactivity": "0" }, { "effective_balance": "30000000000", "head": "199149", "target": "199149", "source": "199149", "inclusion_delay": "293484", "inactivity": "0" }, { "effective_balance": "31000000000", "head": "205787", "target": "205787", "source": "205787", "inclusion_delay": "303267", "inactivity": "0" }, { "effective_balance": "32000000000", "head": "212426", "target": "212426", "source": "212426", "inclusion_delay": "313050", "inactivity": "0" } ], "total_rewards": [ { "validator_index": "0", "head": "212426", "target": "212426", "source": "212426", "inclusion_delay": "313050", "inactivity": "0" }, { "validator_index": "32", "head": "212426", "target": "212426", "source": "212426", "inclusion_delay": "313050", "inactivity": "0" }, { "validator_index": "63", "head": "-357771", "target": "-357771", "source": "-357771", "inclusion_delay": "0", "inactivity": "0" } ] } ``` </details>
2023-07-18 01:48:40 +00:00
#[tokio::test]
async fn test_verify_attestation_rewards_base_subset_only() {
let harness = get_harness(E::default_spec());
// epoch 0 (N), only two thirds of validators vote.
let two_thirds = (VALIDATOR_COUNT / 3) * 2;
let two_thirds_validators: Vec<usize> = (0..two_thirds).collect();
harness
.extend_chain(
E::slots_per_epoch() as usize,
BlockStrategy::OnCanonicalHead,
AttestationStrategy::SomeValidators(two_thirds_validators),
)
.await;
// a small subset of validators to compute attestation rewards for
let validators_subset = [0, VALIDATOR_COUNT / 2, VALIDATOR_COUNT - 1];
// capture balances before transitioning to N + 2
let initial_balances = get_validator_balances(harness.get_current_state(), &validators_subset);
// extend slots to beginning of epoch N + 2
harness.extend_slots(E::slots_per_epoch() as usize).await;
let validators_subset_ids: Vec<ValidatorId> = validators_subset
.into_iter()
.map(|idx| ValidatorId::Index(idx as u64))
.collect();
// compute reward deltas for the subset of validators in epoch N
let StandardAttestationRewards {
ideal_rewards: _,
total_rewards,
} = harness
.chain
.compute_attestation_rewards(Epoch::new(0), validators_subset_ids)
.unwrap();
// apply attestation rewards to initial balances
let expected_balances = apply_attestation_rewards(&initial_balances, total_rewards);
// verify expected balances against actual balances
let balances = get_validator_balances(harness.get_current_state(), &validators_subset);
assert_eq!(expected_balances, balances);
}
/// Apply a vec of `TotalAttestationRewards` to initial balances, and return
fn apply_attestation_rewards(
initial_balances: &[u64],
attestation_rewards: Vec<TotalAttestationRewards>,
) -> Vec<u64> {
initial_balances
.iter()
.zip(attestation_rewards)
.map(|(&initial_balance, rewards)| {
let expected_balance = initial_balance as i64
+ rewards.head
+ rewards.source
+ rewards.target
+ rewards.inclusion_delay.map(|q| q.value).unwrap_or(0) as i64
+ rewards.inactivity;
expected_balance as u64
})
.collect::<Vec<u64>>()
}
fn get_validator_balances(state: BeaconState<E>, validators: &[usize]) -> Vec<u64> {
validators
.iter()
.flat_map(|&id| {
state
.balances()
.get(id)
.cloned()
.ok_or(BeaconStateError::BalancesOutOfBounds(id))
})
.collect()
}
fn apply_beacon_block_rewards(
proposal_rewards_map: &HashMap<u64, u64>,
expected_balances: Vec<u64>,
) -> Vec<u64> {
let calculated_balances = expected_balances
.iter()
.enumerate()
.map(|(i, balance)| balance + proposal_rewards_map.get(&(i as u64)).unwrap_or(&0u64))
.collect();
calculated_balances
}
fn apply_sync_committee_rewards(
sync_committee_rewards_map: &HashMap<u64, i64>,
expected_balances: Vec<u64>,
) -> Vec<u64> {
let calculated_balances = expected_balances
.iter()
.enumerate()
.map(|(i, balance)| {
(*balance as i64 + sync_committee_rewards_map.get(&(i as u64)).unwrap_or(&0i64))
.unsigned_abs()
})
.collect();
calculated_balances
}