staking multiple nominations

[4meta5]

What does it do?

• The previous design only allowed one validator per nominator. This PR replaces that constraint with a new one that allows T::MaxValidatorsPerNominator validators per nominator, such that each nomination > T::MinNomination and total nominated > T::MinNominatorStk.
• update chainspecs with separate input param stakers: Vec<(AccountId, Option<AccountId>, Balance)>. For validators, it is (AccountId, None, Balance) and for nominators it is (AccountId, Some(ValidatorAccountId), Balance). This logic can be found in the stake pallet genesis build() impl.
• validator candidates may bond more/less
• nominators may bond more/less
• switch_nomination runtime method

What important points reviewers should know?

Here are some high-level docs, which should make the code easier to read/review.

Is there something left for follow-up PRs?

enforcing inflation schedule, which was an original goal of this PR

What alternative implementations were considered?

Are there relevant PRs or issues in other repositories (Substrate, Polkadot, Frontier, Cumulus)?

#179

What value does it bring to the blockchain users?

Checklist

• Does it require a purge of the network? No
• You bumped the runtime version if there are breaking changes in the runtime ?
• Does it require changes in documentation/tutorials ? No

[4meta5]

ValidatorStatus used to be generic over the inner variant of Leaving, but it was only ever instantiated with the RoundIndex type. So, the generic was removed and the inner variant of ValidatorStatus::Leaving was defined to always be RoundIndex. Earlier in this PR, something similar was done to remove RoundIndex as a generic parameter from CandidateState.

[4meta5]

This changes the default fee for validators registered at genesis to 0%. It was previously 2%.

We could also add a DefaultValidatorFee config constant and use that here with the old join_candidates method. Then, there's no need to create the join_candidates_no_fee runtime method, but it does add a config constant and there are a few already.

[4meta5]

Never mind, changed this so didn't add a new runtime method. The default fee was still changed to 0% for validators registered at genesis. is this a suitable default?

[4meta5]

This sets the minimum nomination as equal to MinNominatorStk = 5 * GLMR, but it could be set at some value less than MinNominatorStk. These limits exist to prevent trivial nominations from abusing runtime storage.

[4meta5]

• update polkadot-js types based on changes to Rust types

[JoshOrndorff]

The notes are awesome. Can we get those as doc comments or something?

What does it mean to go offline?

Why do we need seperate bond_more (same question about bond_less) methods for validators and nominators? Maybe you could have a single fn bond_more(origin, candidate, more: BalanceOf<T>) and if the candidate is the same as the caller then treat it as a candidate bonding more. Or use an option like you did for the genesis config.

Design question: Can one account be a validator and also a nominator? If so can the account self nominate?

Concern: This pallet just ticked past 1k lines code, and some parts are deeply nested. I'm reassured by the thorough test coverage. But I wonder how this footprint compares to eg pallet staking.

[4meta5]

The notes are awesome. Can we get those as doc comments or something?

Thanks, the reason I made those notes is because I had stuff that needed to be communicated but didn't know where to put the relevant doc comment. For example, the discussion of OrderedSet usage because it's used in a few places. I'll look over it again and see if I can find better places to put some of the information. LMK if you have any suggestions.

What does it mean to go offline?

Leave the pool of candidates so cannot be selected as a validator, but still bonded and can rejoin the candidate pool by calling go_online. It also changes their status to offline so any changes to the nominations or the validator's bond do not require updating the CandidatePool. This is why there are a few places with code like

if candidate.is_active() {
 Self::update_active_candidate()
}

Design question: Can one account be a validator and also a nominator? If so can the account self nominate?

The validator and nominator roles are separate and no account is allowed to occupy both roles. So, when an account tries to join_candidates, there are two checks that (1) it isn't a nominator and (2) it isn't a validator candidate. For join_nominators, it has the same two checks.

So no validator can self-nominate. The validator has to be bonded > T::MinValidatorStk, but that value is stored separately from its nominations. It is the same as a self-nomination in practice, but it does not use the nominator runtime methods.

Why do we need seperate bond_more (same question about bond_less) methods for validators and nominators? Maybe you could have a single fn bond_more(origin, candidate, more: BalanceOf) and if the candidate is the same as the caller then treat it as a candidate bonding more. Or use an option like you did for the genesis config.

I could make this change, but I like having separate runtime methods for validator candidates and nominators because the storage items are separate.

Also I generally try to avoid using Options as input params on runtime methods when it splits the path inside the runtime method and there is not a lot of shared logic between the branches. This changes

fn validator_bond_more() {
   _validator_bond_more()
}

fn nominator_bond_more() {
   _nominator_bond_more()
}

to this

fn bond_more(Option) {
 if let Some(_) = Option {
    _nominator_bond_more()
  } else {
    _validator_bond_more()
  }
}

Because they have separate storage items, the branches don't share any logic so this change makes it more nested in order to combine them into a single method.

Concern: This pallet just ticked past 1k lines code, and some parts are deeply nested. I'm reassured by the thorough test coverage. But I wonder how this footprint compares to eg pallet staking.

Still ~2500 loc less than pallet-staking/src/lib.rs and that doesn't count the other files in that pallet.

To implement support for multiple nominations, I added another Vec that has to be iterated over whenever we add nominations, remove nominations, or change nominations. I think it's worth it, but agree that surface area has increased.

[JoshOrndorff]

Thanks for the answers. That all makes good sense to me.

I think it makes sense to do another learning session where the team gets to stake and validate and nominate so we can develop a stronger intuition for the various code paths. This pallet is really nice.

[JoshOrndorff]

LGTM