[Staking] Allow nominators to be non-slashable and fast unbondable

[Ank4n]

Context

We want to make nominators unslashable (configurable via a storage set), and once they are unslashable, they can also unbond and withdraw in 2 era instead of the full BondingDuration (28 eras).

Storage Changes

• AreNominatorsSlashable: StorageValue<bool> (default: true): Runtime-configurable flag. Made this a storage value (not a config constant) so it can be enabled together with MinValidatorBond and MinCommission via set_staking_configs.
• ErasNominatorsSlashable: StorageMap<EraIndex, bool> (default: true): Per-era snapshot of slashability setting. This ensures offences are processed with the rules that were in effect at the time of the offence, not the current rules. Cleaned up automatically for eras outside bonding window.
• LastValidatorEra to track if a staker was a validator in a recent era and hence needs to follow full unbonding time. Does not need migration as long as we disable nominator slash (in other words: reduce their unbond time) at least one era after these changes are applied.

Slashing logic

• Added process_offence_validator_only as a separate code path instead of overloading the same function. See process_offence_for_era in substrate/frame/staking-async/src/slashing.rs.
• We might want to remove nominator slashing code completely at some point.

Unbonding logic:

• Introduce new config constant NominatorFastUnbondDuration that determines the fast unbond duration (recommended value: 2 eras) when nominators are not slashable.
• Added nominator_bonding_duration() to StakingInterface trait (returns NominatorFastUnbondDuration era when not slashable, full unbond duration otherwise).
• Nomination pools now use nominator_bonding_duration(), so pool members also benefit from fast unbonding.
• Ported auto-chill on full unbond from pallet-staking (PR Full Unbond in Staking #3811) to prevent InsufficientBond errors.
• Nominators unbonding the era before the nominators become unslashable will still have 28 days of unbonding.

Era pruning:

• Moved pruning of ValidatorSlashInEra as well as ErasNominatorsSlashable in lazy pruning. This has a minor (I believe acceptable) side effect that they will be cleaned up in 84 eras instead of 28 eras.

---

TODO

• Ensure delegator slash works correctly (nomination pool).
• Ensure pool members can unbond in 1 day as well.
• Benchmark update.
• Document how all three can be changed in one go: MinCommission, MinValidatorBond, and AreNominatorsSlashable.
• Regenerate weight
• Make nominator unbonding time configurable and set it to 2 eras.
• Refactor compute slash to avoid calling slash_nominator completely.

[burdges]

We do need minimum comission and maybe minimum self stake before that.

[sigurpol]

We do need minimum comission and maybe minimum self stake before that.

Yes, this is the plan: deliver both min self stake + min commission (and support for external self-stake) together with changes for nominators (no slashing, 1-day unbonding) for March, together with the new issuance curve.

[Ank4n]

rename the bonding_duration in StakingInterface to validator_bonding_duration.

[sigurpol]

I wouldn't mind renaming BondingDuration to ValidatorBondingDuration, as we discussed, but it's not a hill I'm willing to die on 😅

[sigurpol]

Whole logic looks solid 🚀
Just some questions / suggestions.
I still have to go through the tests

[sigurpol]

we discussed already but I believe it's a bit cleaner if this method just set the value for the given era and no other side effects. I believe cleaning up fits very well in the lazy era pruning so I would just add something like

 pub(crate) fn clear_era_information(era_index: EraIndex) {
      // ... existing clears ...
      ErasNominatorsSlashable::<T>::remove(era_index);
  }

Ok, we might end up keeping information for 84 eras and not 28 - but it's 1 bool per era so very neglibile

[Ank4n]

But now we use the lazy pruning right? I looked at it and it seems we are having individual steps for each storage, but I think its a bit overkill to break down single storage kills into different steps. To align with current approach, I can do the same, but wondering if you think we should refactor and combine single storage kills in one step?

[sigurpol]

yes I believe it would definitely make sense to combine all the cheap operation like single storage kills in a single step!!!!

[Ank4n]

Some findings

• Lazy pruning works only with history depth (84 eras) based era cleanups.
• There is an existing era based storage (ValidatorsSlashInEra cleaned up every 28 eras) that is not lazily cleaned, and is unbounded (though implicitly bounded by max validators, or if we assume 67% honest validators then 33% of validator set size).

I have few options:

• Keep bonded era (28 eras) based cleanups together (ValidatorSlashInEra + ErasNominatorsSlashable + BondedEras) and eager.
• Move bonded era cleanups also along with lazy pruning that cleans it up 84 eras (I am leaning towards this).
• Create another separate pruning mechanism for bonded era based storages.

@sigurpol any thoughts?

[sigurpol]

Move bonded era cleanups also along with lazy pruning that cleans it up 84 eras (I am leaning towards this).

this seems the cleanest solution imho

[sigurpol]

I wouldn't mind renaming BondingDuration to ValidatorBondingDuration, as we discussed, but it's not a hill I'm willing to die on 😅

[sigurpol]

feedback as outcome of the discussion with @Ank4n , @michalisFr , Jonas: let's say we are at era 100 and planning for 101. Snapshot for 101 is taken at session 1 of era 100. If you (nominator) unbond at Era 100 no matter when, then you are exposed for Era 101 and therefore unlock should happen at era 102 so at most 2d after the unbond. Basically you are always unlocked at the end of next era (so at era 102 in the example).

[kianenigma]

Perfectly done! thanks for bearing with my questions, in the main one I was hallucinating :)

[paritytech-release-backport-bot]

Created backport PR for stable2512:

• [stable2512] Backport #10502 #10810 with remaining conflicts!

Please cherry-pick the changes locally and resolve any conflicts.

git fetch origin backport-10502-to-stable2512
git worktree add --checkout .worktree/backport-10502-to-stable2512 backport-10502-to-stable2512
cd .worktree/backport-10502-to-stable2512
git reset --hard HEAD^
git cherry-pick -x b3bfba618e98f2aa10ee8d4233a15c1e09fef50f
git push --force-with-lease