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

[Staking] Bounded Slashing: Paginated Offence Processing & Slash Application #7424

Merged
merged 212 commits into from
Feb 17, 2025

Conversation

Ank4n
Copy link
Contributor

@Ank4n Ank4n commented Feb 3, 2025

closes #3610.

helps #6344, but need to migrate storage Offences::Reports before we can remove exposure dependency in RC pallets.

replaces #6788.

Context

Slashing in staking is unbounded currently, which is a major blocker until staking can move to a parachain (AH).

Current Slashing Process (Unbounded)

  1. Offence Reported

    • Offences include multiple validators, each with potentially large exposure pages.
    • Slashes are computed immediately and scheduled for application after 28 eras.
  2. Slash Applied

    • All unapplied slashes are executed in one block at the start of the 28th era. This is an unbounded operation.

Proposed Slashing Process (Bounded)

  1. Offence Queueing

    • Offences are queued after basic sanity checks.
  2. Paged Offence Processing (Computing Slash)

    • Slashes are computed one validator exposure page at a time.
    • Unapplied slashes are stored in a double map:
      • Key 1 (k1): EraIndex
      • Key 2 (k2): (Validator, SlashFraction, PageIndex) — a unique identifier for each slash page
  3. Paged Slash Application

    • Slashes are applied one page at a time across multiple blocks.
    • Slash application starts at the 27th era (one era earlier than before) to ensure all slashes are applied before stakers can unbond (which starts from era 28 onwards).

Worst-Case Block Calculation for Slash Application

Polkadot:

  • 1 era = 24 hours, 1 block = 6s14,400 blocks/era
  • On parachains (12s blocks) → 7,200 blocks/era

Kusama:

  • 1 era = 6 hours, 1 block = 6s3,600 blocks/era
  • On parachains (12s blocks) → 1,800 blocks/era

Worst-Case Assumptions:

  • Total stakers: 40,000 nominators, 1000 validators. (Polkadot currently has ~23k nominators and 500 validators)
  • Max slashed: 50% so 20k nominators, 250 validators.
  • Page size: Validators with multiple page: (512 + 1)/2 = 256 , Validators with single page: 1

Calculation:

There might be a more accurate way to calculate this worst-case number, and this estimate could be significantly higher than necessary, but it shouldn’t exceed this value.

Blocks needed: 250 + 20k/256 = ~330 blocks.

Potential Improvement:

  • Consider adding an Offchain Worker (OCW) task to further optimize slash application in future updates.
  • Dynamically batch unapplied slashes based on number of nominators in the page, or process until reserved weight limit is exhausted.

Summary of Changes

Storage

  • New:

    • OffenceQueue (StorageDoubleMap)
      • K1: Era
      • K2: Offending validator account
      • V: OffenceRecord
    • OffenceQueueEras (StorageValue)
      • V: BoundedVec<EraIndex, BoundingDuration>
    • ProcessingOffence (StorageValue)
      • V: (Era, offending validator account, OffenceRecord)
  • Changed:

    • UnappliedSlashes:
      • Old: StorageMap<K -> Era, V -> Vec<UnappliedSlash>>
      • New: StorageDoubleMap<K1 -> Era, K2 -> (validator_acc, perbill, page_index), V -> UnappliedSlash>

Events

  • New:
    • SlashComputed { offence_era, slash_era, offender, page }
    • SlashCancelled { slash_era, slash_key, payout }

Error

  • Changed:
    • InvalidSlashIndex → Renamed to InvalidSlashRecord
  • Removed:
    • NotSortedAndUnique
  • Added:
    • EraNotStarted

Call

  • Changed:
    • cancel_deferred_slash(era, slash_indices: Vec<u32>)
      → Now takes Vec<(validator_acc, slash_fraction, page_index)>
  • New:
    • apply_slash(slash_era, slash_key: (validator_acc, slash_fraction, page_index))

Runtime Config

  • FullIdentification is now set to a unit type (()) / null identity,
    replacing the previous exposure type for all runtimes using pallet_session::historical.

TODO

  • Fixed broken CancelDeferredSlashes.
  • Ensure on_offence called only with validator account for identification everywhere.
  • Ensure we never need to read full exposure.
  • Tests for multi block processing and application of slash.
  • Migrate UnappliedSlashes
  • Bench (crude, needs proper bench as followup)
    • on_offence()
    • process_offence()
    • apply_slash()

Followups (tracker link)

  • OCW task to process offence + apply slashes.
  • Minimum time for governance to cancel deferred slash.
  • Allow root or staking admin to add a custom slash.
  • Test HistoricalSession proof works fine with eras before removing exposure as full identity.
  • Properly bench offence processing and slashing.
  • Handle Offences::Reports migration when removing validator exposure as identity.

gpestana and others added 30 commits October 13, 2024 15:34
Base automatically changed from kiz-multi-block-eletion to master February 15, 2025 00:17
@github-actions github-actions bot requested review from a team as code owners February 15, 2025 00:17
@paritytech-workflow-stopper
Copy link

All GitHub workflows were cancelled due to failure one of the required jobs.
Failed workflow url: https://github.com/paritytech/polkadot-sdk/actions/runs/13348574244
Failed job name: test-linux-stable

@Ank4n Ank4n changed the title [Staking] Make Slashing Bounded with Paged Processing [Staking] Bounded Slashing: Paginated Offence Processing & Slash Application Feb 16, 2025
@paritytech-review-bot paritytech-review-bot bot requested a review from a team February 17, 2025 15:24
Copy link
Contributor

@seadanda seadanda left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

High level LGTM

@seadanda seadanda added this pull request to the merge queue Feb 17, 2025
Merged via the queue into master with commit dda2cb5 Feb 17, 2025
222 of 234 checks passed
@seadanda seadanda deleted the ankan/paged-slashing branch February 17, 2025 23:40
clangenb pushed a commit to clangenb/polkadot-sdk that referenced this pull request Feb 19, 2025
## Multi Block Election Pallet

This PR adds the first iteration of the multi-block staking pallet. 

From this point onwards, the staking and its election provider pallets
are being customized to work in AssetHub. While usage in solo-chains is
still possible, it is not longer the main focus of this pallet. For a
safer usage, please fork and user an older version of this pallet.

---

## Replaces

- [x] paritytech#6034 
- [x] paritytech#5272

## Related PRs: 

- [x] paritytech#7483
- [ ] paritytech#7357
- [ ] paritytech#7424
- [ ] paritytech/polkadot-staking-miner#955

This branch can be periodically merged into
paritytech#7358 ->
paritytech#6996

## TODOs: 

- [x] rebase to master 
- Benchmarking for staking critical path
  - [x] snapshot
  - [x] election result
- Benchmarking for EPMB critical path
  - [x] snapshot
  - [x] verification
  - [x] submission
  - [x] unsigned submission
  - [ ] election results fetching
- [ ] Fix deletion weights. Either of
  - [ ] Garbage collector + lazy removal of all paged storage items
  - [ ] Confirm that deletion is small PoV footprint.
- [ ] Move election prediction to be push based. @tdimitrov 
- [ ] integrity checks for bounds 
- [ ] Properly benchmark this as a part of CI -- for now I will remove
them as they are too slow
- [x] add try-state to all pallets
- [x] Staking to allow genesis dev accounts to be created internally
- [x] Decouple miner config so @niklasad1 can work on the miner
72841b7
- [x] duplicate snapshot page reported by @niklasad1 
- [ ] paritytech#6520 or equivalent
-- during snapshot, `VoterList` must be locked
- [ ] Move target snapshot to a separate block

---------

Co-authored-by: Gonçalo Pestana <[email protected]>
Co-authored-by: Ankan <[email protected]>
Co-authored-by: command-bot <>
Co-authored-by: Guillaume Thiolliere <[email protected]>
Co-authored-by: Giuseppe Re <[email protected]>
Co-authored-by: cmd[bot] <41898282+github-actions[bot]@users.noreply.github.com>
clangenb pushed a commit to clangenb/polkadot-sdk that referenced this pull request Feb 19, 2025
…ication (paritytech#7424)

closes paritytech#3610.

helps paritytech#6344, but need
to migrate storage `Offences::Reports` before we can remove exposure
dependency in RC pallets.

replaces paritytech#6788.

## Context  
Slashing in staking is unbounded currently, which is a major blocker
until staking can move to a parachain (AH).

### Current Slashing Process (Unbounded)  

1. **Offence Reported**  
- Offences include multiple validators, each with potentially large
exposure pages.
- Slashes are **computed immediately** and scheduled for application
after **28 eras**.

2. **Slash Applied**  
- All unapplied slashes are executed in **one block** at the start of
the **28th era**. This is an **unbounded operation**.


### Proposed Slashing Process (Bounded)  

1. **Offence Queueing**  
   - Offences are **queued** after basic sanity checks.  

2. **Paged Offence Processing (Computing Slash)**  
   - Slashes are **computed one validator exposure page at a time**.  
   - **Unapplied slashes** are stored in a **double map**:  
     - **Key 1 (k1):** `EraIndex`  
- **Key 2 (k2):** `(Validator, SlashFraction, PageIndex)` — a unique
identifier for each slash page

3. **Paged Slash Application**  
- Slashes are **applied one page at a time** across multiple blocks.
- Slash application starts at the **27th era** (one era earlier than
before) to ensure all slashes are applied **before stakers can unbond**
(which starts from era 28 onwards).

---

## Worst-Case Block Calculation for Slash Application  

### Polkadot:  
- **1 era = 24 hours**, **1 block = 6s** → **14,400 blocks/era**  
- On parachains (**12s blocks**) → **7,200 blocks/era**  

### Kusama:  
- **1 era = 6 hours**, **1 block = 6s** → **3,600 blocks/era**  
- On parachains (**12s blocks**) → **1,800 blocks/era**  

### Worst-Case Assumptions:  
- **Total stakers:** 40,000 nominators, 1000 validators. (Polkadot
currently has ~23k nominators and 500 validators)
- **Max slashed:** 50% so 20k nominators, 250 validators.  
- **Page size:** Validators with multiple page: (512 + 1)/2 = 256 ,
Validators with single page: 1

### Calculation:  
There might be a more accurate way to calculate this worst-case number,
and this estimate could be significantly higher than necessary, but it
shouldn’t exceed this value.

Blocks needed: 250 + 20k/256 = ~330 blocks.

##  *Potential Improvement:*  
- Consider adding an **Offchain Worker (OCW)** task to further optimize
slash application in future updates.
- Dynamically batch unapplied slashes based on number of nominators in
the page, or process until reserved weight limit is exhausted.

----
## Summary of Changes  

### Storage  
- **New:**  
  - `OffenceQueue` *(StorageDoubleMap)*  
    - **K1:** Era  
    - **K2:** Offending validator account  
    - **V:** `OffenceRecord`  
  - `OffenceQueueEras` *(StorageValue)*  
    - **V:** `BoundedVec<EraIndex, BoundingDuration>`  
  - `ProcessingOffence` *(StorageValue)*  
    - **V:** `(Era, offending validator account, OffenceRecord)`  

- **Changed:**  
  - `UnappliedSlashes`:  
    - **Old:** `StorageMap<K -> Era, V -> Vec<UnappliedSlash>>`  
- **New:** `StorageDoubleMap<K1 -> Era, K2 -> (validator_acc, perbill,
page_index), V -> UnappliedSlash>`

### Events  
- **New:**  
  - `SlashComputed { offence_era, slash_era, offender, page }`  
  - `SlashCancelled { slash_era, slash_key, payout }`  

### Error  
- **Changed:**  
  - `InvalidSlashIndex` → Renamed to `InvalidSlashRecord`  
- **Removed:**  
  - `NotSortedAndUnique`  
- **Added:**  
  - `EraNotStarted`  

### Call  
- **Changed:**  
  - `cancel_deferred_slash(era, slash_indices: Vec<u32>)`  
    → Now takes `Vec<(validator_acc, slash_fraction, page_index)>`  
- **New:**  
- `apply_slash(slash_era, slash_key: (validator_acc, slash_fraction,
page_index))`

### Runtime Config  
- `FullIdentification` is now set to a unit type (`()`) / null identity,
replacing the previous exposure type for all runtimes using
`pallet_session::historical`.

## TODO
- [x] Fixed broken `CancelDeferredSlashes`.
- [x] Ensure on_offence called only with validator account for
identification everywhere.
- [ ] Ensure we never need to read full exposure.
- [x] Tests for multi block processing and application of slash.
- [x] Migrate UnappliedSlashes 
- [x] Bench (crude, needs proper bench as followup)
  - [x] on_offence()
  - [x] process_offence()
  - [x] apply_slash()
 
 
## Followups (tracker
[link](paritytech#7596))
- [ ] OCW task to process offence + apply slashes.
- [ ] Minimum time for governance to cancel deferred slash.
- [ ] Allow root or staking admin to add a custom slash.
- [ ] Test HistoricalSession proof works fine with eras before removing
exposure as full identity.
- [ ] Properly bench offence processing and slashing.
- [ ] Handle Offences::Reports migration when removing validator
exposure as identity.

---------

Co-authored-by: Gonçalo Pestana <[email protected]>
Co-authored-by: command-bot <>
Co-authored-by: Kian Paimani <[email protected]>
Co-authored-by: Guillaume Thiolliere <[email protected]>
Co-authored-by: kianenigma <[email protected]>
Co-authored-by: Giuseppe Re <[email protected]>
Co-authored-by: cmd[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
I4-refactor Code needs refactoring. T2-pallets This PR/Issue is related to a particular pallet.
Projects
Status: Done
Status: In progress
Development

Successfully merging this pull request may close these issues.

[NPoS] Pagify slashing
7 participants