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

BEP-283: Segmented History Data Maintenance #283

Open
wants to merge 4 commits into
base: master
Choose a base branch
from

Conversation

Mercybudda
Copy link

@Mercybudda Mercybudda commented Sep 7, 2023

  BEP: 283
  Title: Segmented History Block Data Maintenance
  Status: Draft
  Type: Standards
  Created: 2023-09-07

BEP-283: Segmented History Block Maintenance

1. Summary

This BEP proposes a practical solution to address the issue of increasing history data storage on the BSC(BNB Smart Chain) for full nodes, so they only need to maintain a limited range of blocks. With the introduction of a new synchronization protocol and storage protocol, nodes can synchronize from a specific checkpoint and skip historical blocks. This eliminates the mandatory requirement to keep historical blocks.

2. Motivation

The history block data includes block header, body and receipt, they are useless to execute the latest blocks. A recent storage profile on Jul-02-2023 shows that the size of history block has reached ~1288 GB, which makes it a big burden to run a BSC full node. As BSC keeps generating new blocks, the history block data also keeps growing.
Actually the history block data can be maintained by some archive node or by a DA layer, a full node does not need to keep a full copy of the history block. But simply deleting the history block data is also not acceptable, it will make the P2P network in chaos and hard to sync.
We need to make a rule so that full nodes only need to keep a recent period of blocks, like several months, it would be better to keep the bounded history block data size within 200GB. It could reduce both node’s storage pressure and the network traffic cost.

3. Specification

Parameter Value Description
BoundStartBlock 31,268,530 The starting block height of the first segment, block 31,268,530 was produced on Aug-29-2023, 3 years after the BSC genesis block.
HistorySegmentLength 2,592,000 assume 1 block for every 3 second, 2,592,000 blocks will be produced in 90 days.

3.1 General Workflow

The history block data will be divided into several segments, the 1st segment is segment_0, which is from genesis to BoundStartBlock-1, then following segments will have the same length: HistorySegmentLength.

HistorySegment_0 = [0, BoundStartBlock)
HistorySegment_1 = [BoundStartBlock, BoundStartBlock+HistorySegmentLength)
...
HistorySegment_N = [BoundStartBlock+(N-1)*HistorySegmentLength, BoundStartBlock+N*HistorySegmentLength)

The BSC node only needs to maintain the latest 2 segments, in case of block reorg, the current segment number is calculated based on the “finalized” block.

ImmutabilityThreshold = 90000

func GetFinalizedBlock() {
  blockByAttestation := GetFinalizedByAttestation()
  blockByThreshold := GetCurrentBlockNumber() - ImmutabilityThreshold
  return max(blockByAttestation, blockByThreshold)
}

func GetCurrentSegmentIndex(blockNum int) {
  if blockNum < BoundStartBlock {
    return 0
  }
  boundBlocks = blockNum - BoundStartBlock
  return (boundBlocks/HistorySegmentLength) + 1
}

finalizedBlock = GetFinalizedBlock()
segIndex = GetCurrentSegmentIndex(finalizedBlock)
if  segIndex > = 2 {
  // the segments before (segIndex - 2) can be pruned
}

3.2 Prune

The offline block prune would need to be boundary aligned and leave the most recent 2 segments. The most recent segment is determined by the finalized block, since FastFinality is to be enabled, we can take use of the feature to determine the finalized block and then get the current segment index.

3.2.1 Prune details

Since we introduced segment definition, the historical block data will be organized on a segment-by-segment basis within the file directory, making it easy to prune the block data by deleting the corresponding file folder.

if segIndex > 2 {
  // delete the file folder of segIndex0 to segIndex - 2
  // os.RemoveAll(segIndex0..segIndex-2)
}

3.3 Node Sync

It would be difficult to sync from genesis, since most nodes may choose to not preserve these old blocks, but it is still possible as long as some nodes still keep the whole history data.
There could have 2 approaches to do sync after this proposal
directly download a snapshot of a recent state, from a snapshot service provider or DA layer like GreenField
Segment based snap sync: user can take the boundary blockhash as the new GenesisHash and start from it directly.

3.4 Data Availability

Some of the nodes like archive nodes would keep maintaining the whole history data.
And meanwhile, could take use of a DA layer, like greenfield to make sure the whole block data is available.

4. Rationale

4.1 BoundStartBlock & HistorySegmentLength

As the current history block data is already very large, we prefer to enable this proposal faster, so
BoundStartBlock is still more than 1 month ahead, it could be an acceptable date.
HistorySegmentLength, we did a profile, for the past 6 months(Jan-2023 to July 2023), ~1.2GB history data was generated per day on average. But it is somehow low traffic during this period. Traffic volume could be 3 times if the bull market starts, that is ~3.6GB per day. To keep the historical data size within 500GB in the bull market and 200GB in the bear market, 90 days could be a reasonable value.

4.2 Why Segment The History Data

No Big fundamental differences to the EIP-4444 proposal
Pros: aligned
easier to determine the current living blocks?
easier to provide the snapshot download service?

5. Forward Compatibility

5.1 Portal Network

Portal network is a hot topic to solve the storage pressure, once it is ready, it is possible that portal network can replace this proposal if it has a more applicable solution.

6. Backward Compatibility

6.1 Archive Node

If you run an archive node, you can just keep all the history block data, no impact to its business.

6.2 RPC API

If users query history block data that have been pruned, could return a new error code to show it is expired and removed.

License

The content is licensed under the CC0 license.

@Nouri11190
Copy link

Compare the changes and pull request

@Nouri11190 Nouri11190 mentioned this pull request Oct 6, 2023
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

5 participants