Bonding mechanism

Epoch

An epoch is a range of blocks or time that is defined by the base ledger and made available to the PoS system. This document assumes that epochs are identified by consecutive natural numbers. All the data relevant to PoS are associated with epochs.

Epoched data

Epoched data are data associated with a specific epoch that are set in advance. The data relevant to the PoS system in the ledger's state are epoched. Each data can be uniquely identified. These are:

Changes to the epoched data do not take effect immediately. Instead, changes in epoch n are queued to take effect in the epoch n + pipeline_length for most cases and n + unboding_length for unbonding actions. Should the same validator's data or same bonds (i.e. with the same identity) be updated more than once in the same epoch, the later update overrides the previously queued-up update. For bonds, the token amounts are added up. Once the epoch n has ended, the queued-up updates for epoch n + pipeline_length are final and the values become immutable.

Entities

• Validator: An account with a public consensus key, which may participate in producing blocks and governance activities. A validator may not also be a delegator.
• Delegator: An account that delegates some tokens to a validator. A delegator may not also be a validator.

Additionally, any account may submit evidence for a slashable misbehaviour.

Validator

A validator must have a public consensus key. Additionally, it may also specify optional metadata fields (TBA).

A validator may be in one of the following states:

• inactive: A validator is not being considered for block creation and cannot receive any new delegations.
• candidate: A validator is considered for block creation and can receive delegations.

For each validator (in any state), the system also tracks total bonded tokens as a sum of the tokens in their self-bonds and delegated bonds. The total bonded tokens determine their voting voting power by multiplication by the votes_per_token parameter. The voting power is used for validator selection for block creation and is used in governance related activities.

Validator actions

• become validator: Any account that is not a validator already and that doesn't have any delegations may request to become a validator. It is required to provide a public consensus key and staking reward address. For the action applied in epoch n, the validator's state will be set to candidate for epoch n + pipeline_length and the consensus key is set for epoch n + pipeline_length.
• deactivate: Only a validator whose state at or before the pipeline_length offset is candidate account may deactivate. For this action applied in epoch n, the validator's account is set to become inactive in the epoch n + pipeline_length.
• reactivate: Only an inactive validator may reactivate. Similarly to become validator action, for this action applied in epoch n, the validator's state will be set to candidate for epoch n + pipeline_length.
• self-bond: A validator may lock-up tokens into a bond only for its own validator's address.
• unbond: Any self-bonded tokens may be partially or fully unbonded.
• withdraw unbonds: Unbonded tokens may be withdrawn in or after the unbond's epoch.
• change consensus key: Set the new consensus key. When applied in epoch n, the key is set for epoch n + pipeline_length.

Active validator set

From all the candidate validators, in each epoch the ones with the most voting power limited up to the max_validator_slots parameter are selected for the active validator set. The active validator set selected in epoch n is set for epoch n + pipeline_length.

Delegator

A delegator may have any number of delegations to any number of validators. Delegations are stored in bonds.

Delegator actions

• delegate: An account which is not a validator may delegate tokens to any number of validators. This will lock-up tokens into a bond.
• undelegate: Any delegated tokens may be partially or fully unbonded.
• withdraw unbonds: Unbonded tokens may be withdrawn in or after the unbond's epoch.

Bonds

A bond locks-up tokens from validators' self-bonding and delegators' delegations. For self-bonding, the source address is equal to the validator's address. Only validators can self-bond. For a bond created from a delegation, the bond's source is the delegator's account.

For each epoch, bonds are uniquely identified by the pair of source and validator's addresses. A bond created in epoch n is written into epoch n + pipeline_length. If there already is a bond in the epoch n + pipeline_length for this pair of source and validator's addresses, its tokens are incremented by the newly bonded amount.

Any bonds created in epoch n increment the bond's validator's total bonded tokens by the bond's token amount and update the voting power for epoch n + pipeline_length.

The tokens put into a bond are immediately deducted from the source account.

Unbond

An unbonding action (validator unbond or delegator undelegate) requested by the bond's source account in epoch n creates an "unbond" with epoch set to n + unbounding_length. We also store the epoch of the bond(s) from which the unbond is created in order to determine if the unbond should be slashed if a fault occurred within the range of bond epoch (inclusive) and unbond epoch (exclusive).

Any unbonds created in epoch n decrements the bond's validator's total bonded tokens by the bond's token amount and update the voting power for epoch n + unbonding_length.

An "unbond" with epoch set to n may be withdrawn by the bond's source address in or any time after the epoch n. Once withdrawn, the unbond is deleted and the tokens are credited to the source account.

Note that unlike bonding and unbonding where token changes are delayed to some future epochs (pipeline or unbonding offset), the token withdrawal applies immediately. This because when the tokens are withdrawable, they are already "unlocked" from the PoS system and do not contribute to voting power.

Staking rewards

Until we have programmable validity predicates, rewards can use the mechanism outlined in the F1 paper, but it should use the exponential model, so that withdrawing rewards more frequently provides no additional benefit (this is a design constraint we should follow in general, we don't want to accidentally encourage transaction spam). This should be written in a way that allows for a natural upgrade to a validator-customisable rewards model (defaulting to this one) if possible.

To a validator who proposed a block, the system rewards tokens based on the block_proposer_reward system parameter and each validator that voted on a block receives block_vote_reward.

Slashing

An important part of the security model of Namada is based on making attacking the system very expensive. To this end, the validator who has bonded stake will be slashed once an offence has been detected.

These are the types of offences:

• Equivocation in consensus
• voting: meaning that a validator has submitted two votes that are confliciting
• block production: a block producer has created two different blocks for the same height
• Invalidity:
• block production: a block producer has produced invalid block
• voting: validators have voted on invalid block

Unavailability is not considered an offense, but a validator who hasn't voted will not receive rewards.

Once an offence has been reported:

1. Kicking out
2. Slashing
• Individual: Once someone has reported an offence it is reviewed by validarors and if confirmed the offender is slashed.
• cubic slashing: escalated slashing

Instead of absolute values, validators' total bonded token amounts and bonds' and unbonds' token amounts are stored as their deltas (i.e. the change of quantity from a previous epoch) to allow distinguishing changes for different epoch, which is essential for determining whether tokens should be slashed. However, because slashes for a fault that occurred in epoch n may only be applied before the beginning of epoch n + unbonding_length, in epoch m we can sum all the deltas of total bonded token amounts and bonds and unbond with the same source and validator for epoch equal or less than m - unboding_length into a single total bonded token amount, single bond and single unbond record. This is to keep the total number of total bonded token amounts for a unique validator and bonds and unbonds for a unique pair of source and validator bound to a maximum number (equal to unbonding_length).

To disincentivize validators misbehaviour in the PoS system a validator may be slashed for any fault that it has done. An evidence of misbehaviour may be submitted by any account for a fault that occurred in epoch n anytime before the beginning of epoch n + unbonding_length.

A valid evidence reduces the validator's total bonded token amount by the slash rate in and before the epoch in which the fault occurred. The validator's voting power must also be adjusted to the slashed total bonded token amount. Additionally, a slash is stored with the misbehaving validator's address and the relevant epoch in which the fault occurred. When an unbond is being withdrawn, we first look-up if any slash occurred within the range of epochs in which these were active and if so, reduce its token amount by the slash rate. Note that bonds and unbonds amounts are not slashed until their tokens are withdrawn.

The invariant is that the sum of amounts that may be withdrawn from a misbehaving validator must always add up to the total bonded token amount.

System parameters

The default values that are relative to epoch duration assume that an epoch last about 24 hours.

• max_validator_slots: Maximum active validators, default 128
• pipeline_len: Pipeline length in number of epochs, default 2 (see https://github.com/cosmos/cosmos-sdk/blob/019444ae4328beaca32f2f8416ee5edbac2ef30b/docs/architecture/adr-039-epoched-staking.md#pipelining-the-epochs)
• unboding_len: Unbonding duration in number of epochs, default 6
• votes_per_token: Used in validators' voting power calculation, default 100‱ (1 voting power unit per 1000 tokens)
• block_proposer_reward: Amount of tokens rewarded to a validator for proposing a block
• block_vote_reward: Amount of tokens rewarded to each validator that voted on a block proposal
• duplicate_vote_slash_rate: Portion of validator's stake that should be slashed on a duplicate vote
• light_client_attack_slash_rate: Portion of validator's stake that should be slashed on a light client attack

Storage

The system parameters are written into the storage to allow for their changes. Additionally, each validator may record a new parameters value under their sub-key that they wish to change to, which would override the systems parameters when more than 2/3 voting power are in agreement on all the parameters values.

The validators' data are keyed by the their addresses, conceptually:

type Validators = HashMap<Address, Validator>;


Epoched data are stored in the following structure:

struct Epoched<Data> {
/// The epoch in which this data was last updated
last_update: Epoch,
/// Dynamically sized vector in which the head is the data for epoch in which
/// the last_update was performed and every consecutive array element is the
/// successor epoch of the predecessor array element. For system parameters,
/// validator's consensus key and state, LENGTH = pipeline_length + 1.
/// For all others, LENGTH = unbonding_length + 1.
data: Vec<Option<Data>>
}


Note that not all epochs will have data set, only the ones in which some changes occurred.

To try to look-up a value for Epoched data with independent values in each epoch (such as the active validator set) in the current epoch n:

1. let index = min(n - last_update, pipeline_length)
2. read the data field at index:
1. if there's a value at index return it
2. else if index == 0, return None
3. else decrement index and repeat this sub-step from 1.

To look-up a value for Epoched data with delta values in the current epoch n:

1. let end = min(n - last_update, pipeline_length) + 1
2. sum all the values that are not None in the 0 .. end range bounded inclusively below and exclusively above

To update a value in Epoched data with independent values in epoch n with value new for epoch m:

1. let shift = min(n - last_update, pipeline_length)
2. if shift == 0:
1. data[m - n] = new
3. else:
1. for i in 0 .. shift range bounded inclusively below and exclusively above, set data[i] = None
2. rotate data left by shift
3. set data[m - n] = new
4. set last_update to the current epoch

To update a value in Epoched data with delta values in epoch n with value delta for epoch m:

1. let shift = min(n - last_update, pipeline_length)
2. if shift == 0:
1. set data[m - n] = data[m - n].map_or_else(delta, |last_delta| last_delta + delta) (add the delta to the previous value, if any, otherwise use the delta as the value)
3. else:
1. let sum to be equal to the sum of all delta values in the i in 0 .. shift range bounded inclusively below and exclusively above and set data[i] = None
2. rotate data left by shift
3. set data[0] = data[0].map_or_else(sum, |last_delta| last_delta + sum)
4. set data[m - n] = delta
5. set last_update to the current epoch

The invariants for updates in both cases are that m - n >= 0 and m - n <= pipeline_length.

For the active validator set, we store all the active and inactive validators separately with their respective voting power:

type VotingPower = u64;

/// Validator's address with its voting power.
#[derive(PartialEq, Eq, PartialOrd, Ord)]
struct WeightedValidator {
/// The voting_power field must be on top, because lexicographic ordering is
/// based on the top-to-bottom declaration order and in the ValidatorSet
/// the WeighedValidators these need to be sorted by the voting_power.
voting_power: VotingPower,
}

struct ValidatorSet {
/// Active validator set with maximum size equal to max_validator_slots
active: BTreeSet<WeightedValidator>,
/// All the other validators that are not active
inactive: BTreeSet<WeightedValidator>,
}

type ValidatorSets = Epoched<ValidatorSet>;

/// The sum of all active and inactive validators' voting power
type TotalVotingPower = Epoched<VotingPower>;


When any validator's voting power changes, we attempt to perform the following update on the ActiveValidatorSet:

1. let validator be the validator's address, power_before and power_after be the voting power before and after the change, respectively
2. let power_delta = power_after - power_before
3. let min_active = active.first() (active validator with lowest voting power)
4. let max_inactive = inactive.last() (inactive validator with greatest voting power)
5. find whether the validator is active, let is_active = power_before >= max_inactive.voting_power
1. if is_active:
1. if power_delta > 0 && power_after > max_inactive.voting_power, update the validator in active set with voting_power = power_after
2. else, remove the validator from active, insert it into inactive and remove max_inactive.address from inactive and insert it into active
2. else (!is_active):
1. if power_delta < 0 && power_after < min_active.voting_power, update the validator in inactive set with voting_power = power_after
2. else, remove the validator from inactive, insert it into active and remove min_active.address from active and insert it into inactive

Within each validator's address space, we store public consensus key, state, total bonded token amount and voting power calculated from the total bonded token amount (even though the voting power is stored in the ValidatorSet, we also need to have the voting_power here because we cannot look it up in the ValidatorSet without iterating the whole set):

struct Validator {
consensus_key: Epoched<PublicKey>,
state: Epoched<ValidatorState>,
total_deltas: Epoched<token::Amount>,
voting_power: Epoched<VotingPower>,
}

enum ValidatorState {
Inactive,
Candidate,
}


The bonds and unbonds are keyed by their identifier:

type Bonds = HashMap<BondId, Epoched<Bond>>;
type Unbonds = HashMap<BondId, Epoched<Unbond>>;

struct BondId {
/// The delegator adddress for delegations, or the same as the validator
}

struct Bond {
/// A key is a the epoch set for the bond. This is used in unbonding, where
// it's needed for slash epoch range check.
deltas: HashMap<Epoch, token::Amount>,
}

struct Unbond {
/// A key is a pair of the epoch of the bond from which a unbond was created
/// the epoch of unboding. This is needed for slash epoch range check.
deltas: HashMap<(Epoch, Epoch), token::Amount>
}


For slashes, we store the epoch and block height at which the fault occurred, slash rate and the slash type:

struct Slash {
epoch: Epoch,
block_height: u64,
/// slash token amount ‱ (per ten thousand)
rate: u8,
r#type: SlashType,
}


Initialization

An initial validator set with self-bonded token amounts must be given on system initialization.

This set is used to pre-compute epochs in the genesis block from epoch 0 to epoch pipeline_length - 1.