Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Tracker issue for cleaning up old non-paged exposure logic in staking pallet #433

Open
2 of 6 tasks
Ank4n opened this issue Dec 29, 2022 · 1 comment · May be fixed by #5986
Open
2 of 6 tasks

Tracker issue for cleaning up old non-paged exposure logic in staking pallet #433

Ank4n opened this issue Dec 29, 2022 · 1 comment · May be fixed by #5986
Assignees

Comments

@Ank4n
Copy link
Contributor

Ank4n commented Dec 29, 2022

This issue is to track merging of paritytech/substrate#13059 and subsequent release of paged rewards on Polkadot. Once enabled, and 84 eras have passed, we can clean up incremental migration and fallback logic from the staking pallet.

84 eras in Polkadot = 84 days.
Deprecated storage items can be cleaned from codebase on or after Era 1504 (approx: July 14, 2024) on Polkadot.

Track

  • Paged rewards enabled in Kusama on Era 6514 (Apr 18).
  • Paged rewards enabled in Polkadot on Era 1420 (Apr 21).

Tasks

  • Remove storage item ErasStakersClipped and ErasStakers.
  • Remove fn EraInfo::is_rewards_claimed_temp.
  • Clean up logic around StakingLedger.legacy_claimed_rewards.
  • BoundedVec everything

Note on bounding:

I did not bound new paged storage items since the exposure page size would decrease in future which is unsafe for bounded vecs. Initially, page size would be 512 and page count = 1, but after running for a while, we should set the page size such that we don't see them reducing further. For example, when we clean up this ticket, we could then bound the page size to 32 and page count to 16. From that point, we should strive to never reduce the page size and page count should only increase further.

@Ank4n Ank4n self-assigned this Dec 29, 2022
@Ank4n Ank4n changed the title Tracker for cleaning up StakingLedger.claimed_rewards Tracker issue for cleaning up non-paged claimed rewards Jan 16, 2023
@Ank4n Ank4n changed the title Tracker issue for cleaning up non-paged claimed rewards Tracker issue for cleaning up old non-paged exposure logic in staking pallet Feb 13, 2023
@juangirini juangirini transferred this issue from paritytech/substrate Aug 24, 2023
Ank4n added a commit that referenced this issue Nov 1, 2023
…1189)

helps #439.
closes #473.

PR link in the older substrate repository:
paritytech/substrate#13498.

# Context
Rewards payout is processed today in a single block and limited to
`MaxNominatorRewardedPerValidator`. This number is currently 512 on both
Kusama and Polkadot.

This PR tries to scale the nominators payout to an unlimited count in a
multi-block fashion. Exposures are stored in pages, with each page
capped to a certain number (`MaxExposurePageSize`). Starting out, this
number would be the same as `MaxNominatorRewardedPerValidator`, but
eventually, this number can be lowered through new runtime upgrades to
limit the rewardeable nominators per dispatched call instruction.

The changes in the PR are backward compatible.

## How payouts would work like after this change
Staking exposes two calls, 1) the existing `payout_stakers` and 2)
`payout_stakers_by_page`.

### payout_stakers
This remains backward compatible with no signature change. If for a
given era a validator has multiple pages, they can call `payout_stakers`
multiple times. The pages are executed in an ascending sequence and the
runtime takes care of preventing double claims.

### payout_stakers_by_page
Very similar to `payout_stakers` but also accepts an extra param
`page_index`. An account can choose to payout rewards only for an
explicitly passed `page_index`.

**Lets look at an example scenario**
Given an active validator on Kusama had 1100 nominators,
`MaxExposurePageSize` set to 512 for Era e. In order to pay out rewards
to all nominators, the caller would need to call `payout_stakers` 3
times.

- `payout_stakers(origin, stash, e)` => will pay the first 512
nominators.
- `payout_stakers(origin, stash, e)` => will pay the second set of 512
nominators.
- `payout_stakers(origin, stash, e)` => will pay the last set of 76
nominators.
...
- `payout_stakers(origin, stash, e)` => calling it the 4th time would
return an error `InvalidPage`.

The above calls can also be replaced by `payout_stakers_by_page` and
passing a `page_index` explicitly.

## Commission note
Validator commission is paid out in chunks across all the pages where
each commission chunk is proportional to the total stake of the current
page. This implies higher the total stake of a page, higher will be the
commission. If all the pages of a validator's single era are paid out,
the sum of commission paid to the validator across all pages should be
equal to what the commission would have been if we had a non-paged
exposure.

### Migration Note
Strictly speaking, we did not need to bump our storage version since
there is no migration of storage in this PR. But it is still useful to
mark a storage upgrade for the following reasons:

- New storage items are introduced in this PR while some older storage
items are deprecated.
- For the next `HistoryDepth` eras, the exposure would be incrementally
migrated to its corresponding paged storage item.
- Runtimes using staking pallet would strictly need to wait at least
`HistoryDepth` eras with current upgraded version (14) for the migration
to complete. At some era `E` such that `E >
era_at_which_V14_gets_into_effect + HistoryDepth`, we will upgrade to
version X which will remove the deprecated storage items.
In other words, it is a strict requirement that E<sub>x</sub> -
E<sub>14</sub> > `HistoryDepth`, where
E<sub>x</sub> = Era at which deprecated storages are removed from
runtime,
E<sub>14</sub> = Era at which runtime is upgraded to version 14.
- For Polkadot and Kusama, there is a [tracker
ticket](#433) to clean
up the deprecated storage items.

### Storage Changes

#### Added
- ErasStakersOverview
- ClaimedRewards
- ErasStakersPaged

#### Deprecated
The following can be cleaned up after 84 eras which is tracked
[here](#433).

- ErasStakers.
- ErasStakersClipped.
- StakingLedger.claimed_rewards, renamed to
StakingLedger.legacy_claimed_rewards.

### Config Changes
- Renamed MaxNominatorRewardedPerValidator to MaxExposurePageSize.

### TODO
- [x] Tracker ticket for cleaning up the old code after 84 eras.
- [x] Add companion.
- [x] Redo benchmarks before merge.
- [x] Add Changelog for pallet_staking.
- [x] Pallet should be configurable to enable/disable paged rewards.
- [x] Commission payouts are distributed across pages.
- [x] Review documentation thoroughly.
- [x] Rename `MaxNominatorRewardedPerValidator` ->
`MaxExposurePageSize`.
- [x] NMap for `ErasStakersPaged`.
- [x] Deprecate ErasStakers.
- [x] Integrity tests.

### Followup issues
[Runtime api for deprecated ErasStakers storage
item](#426)

---------

Co-authored-by: Javier Viola <javier@parity.io>
Co-authored-by: Ross Bulat <ross@parity.io>
Co-authored-by: command-bot <>
bgallois pushed a commit to duniter/duniter-polkadot-sdk that referenced this issue Mar 25, 2024
…aritytech#1189)

helps paritytech#439.
closes paritytech#473.

PR link in the older substrate repository:
paritytech/substrate#13498.

# Context
Rewards payout is processed today in a single block and limited to
`MaxNominatorRewardedPerValidator`. This number is currently 512 on both
Kusama and Polkadot.

This PR tries to scale the nominators payout to an unlimited count in a
multi-block fashion. Exposures are stored in pages, with each page
capped to a certain number (`MaxExposurePageSize`). Starting out, this
number would be the same as `MaxNominatorRewardedPerValidator`, but
eventually, this number can be lowered through new runtime upgrades to
limit the rewardeable nominators per dispatched call instruction.

The changes in the PR are backward compatible.

## How payouts would work like after this change
Staking exposes two calls, 1) the existing `payout_stakers` and 2)
`payout_stakers_by_page`.

### payout_stakers
This remains backward compatible with no signature change. If for a
given era a validator has multiple pages, they can call `payout_stakers`
multiple times. The pages are executed in an ascending sequence and the
runtime takes care of preventing double claims.

### payout_stakers_by_page
Very similar to `payout_stakers` but also accepts an extra param
`page_index`. An account can choose to payout rewards only for an
explicitly passed `page_index`.

**Lets look at an example scenario**
Given an active validator on Kusama had 1100 nominators,
`MaxExposurePageSize` set to 512 for Era e. In order to pay out rewards
to all nominators, the caller would need to call `payout_stakers` 3
times.

- `payout_stakers(origin, stash, e)` => will pay the first 512
nominators.
- `payout_stakers(origin, stash, e)` => will pay the second set of 512
nominators.
- `payout_stakers(origin, stash, e)` => will pay the last set of 76
nominators.
...
- `payout_stakers(origin, stash, e)` => calling it the 4th time would
return an error `InvalidPage`.

The above calls can also be replaced by `payout_stakers_by_page` and
passing a `page_index` explicitly.

## Commission note
Validator commission is paid out in chunks across all the pages where
each commission chunk is proportional to the total stake of the current
page. This implies higher the total stake of a page, higher will be the
commission. If all the pages of a validator's single era are paid out,
the sum of commission paid to the validator across all pages should be
equal to what the commission would have been if we had a non-paged
exposure.

### Migration Note
Strictly speaking, we did not need to bump our storage version since
there is no migration of storage in this PR. But it is still useful to
mark a storage upgrade for the following reasons:

- New storage items are introduced in this PR while some older storage
items are deprecated.
- For the next `HistoryDepth` eras, the exposure would be incrementally
migrated to its corresponding paged storage item.
- Runtimes using staking pallet would strictly need to wait at least
`HistoryDepth` eras with current upgraded version (14) for the migration
to complete. At some era `E` such that `E >
era_at_which_V14_gets_into_effect + HistoryDepth`, we will upgrade to
version X which will remove the deprecated storage items.
In other words, it is a strict requirement that E<sub>x</sub> -
E<sub>14</sub> > `HistoryDepth`, where
E<sub>x</sub> = Era at which deprecated storages are removed from
runtime,
E<sub>14</sub> = Era at which runtime is upgraded to version 14.
- For Polkadot and Kusama, there is a [tracker
ticket](paritytech#433) to clean
up the deprecated storage items.

### Storage Changes

#### Added
- ErasStakersOverview
- ClaimedRewards
- ErasStakersPaged

#### Deprecated
The following can be cleaned up after 84 eras which is tracked
[here](paritytech#433).

- ErasStakers.
- ErasStakersClipped.
- StakingLedger.claimed_rewards, renamed to
StakingLedger.legacy_claimed_rewards.

### Config Changes
- Renamed MaxNominatorRewardedPerValidator to MaxExposurePageSize.

### TODO
- [x] Tracker ticket for cleaning up the old code after 84 eras.
- [x] Add companion.
- [x] Redo benchmarks before merge.
- [x] Add Changelog for pallet_staking.
- [x] Pallet should be configurable to enable/disable paged rewards.
- [x] Commission payouts are distributed across pages.
- [x] Review documentation thoroughly.
- [x] Rename `MaxNominatorRewardedPerValidator` ->
`MaxExposurePageSize`.
- [x] NMap for `ErasStakersPaged`.
- [x] Deprecate ErasStakers.
- [x] Integrity tests.

### Followup issues
[Runtime api for deprecated ErasStakers storage
item](paritytech#426)

---------

Co-authored-by: Javier Viola <javier@parity.io>
Co-authored-by: Ross Bulat <ross@parity.io>
Co-authored-by: command-bot <>
@Ank4n
Copy link
Contributor Author

Ank4n commented Apr 22, 2024

Referenda for upgrade containing paged exposure:

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
Status: ✂️ In progress.
Development

Successfully merging a pull request may close this issue.

1 participant