Skip to content

DIVERGENCE.md

Latest commit

 

History

History
72 lines (52 loc) · 4.38 KB

DIVERGENCE.md

File metadata and controls

72 lines (52 loc) · 4.38 KB

Divergence from Main Branch

This branch (perf/block-optimization) introduces breaking changes to maximize performance in the block/ package. It is not wire-compatible with the main branch.

1. Combined Header+Data Blobs

Main: Headers and data are submitted as separate blobs to separate DA namespaces (HeaderNamespace, DataNamespace). On retrieval, the DA retriever fetches from both namespaces, decodes headers and data independently, then matches them by block height.

This branch: Headers and data are combined into a single blob using a custom binary encoding (common.MarshalBlockBlob/UnmarshalBlockBlob). Each blob contains the proto-encoded header, proto-encoded data, and the envelope signature, separated by length prefixes with a magic number prefix (0x45564E44). Only the HeaderNamespace is used.

Why

  • Eliminates matching overhead (no separate header/data pending maps)
  • Halves DA submission round trips (one blob per block instead of two)
  • Simplifies DA inclusion tracking (single check per block)
  • Removes the DAHeaderEnvelope protobuf wrapper and the separate SignedData protobuf wrapper

2. Custom Binary Blob Encoding

Main: DA blobs use protobuf encoding (DAHeaderEnvelope for headers, SignedData for data). Each involves allocating proto message structs, converting Go types to proto types, and calling proto.Marshal.

This branch: The combined blob wrapper uses a custom binary format: [magic 4B][header_len 4B][header_bytes][data_len 4B][data_bytes][sig_len 4B][sig_bytes]. Individual header and data fields are still proto-encoded internally (hash computation requires it), but the envelope wrapper avoids all proto overhead.

Why

  • Zero allocation for the blob wrapper (direct length-prefixed binary)
  • No proto message pool management for the envelope
  • No ToProto/FromProto conversion for the DA envelope or signed data
  • Simpler and faster encode/decode path

3. P2P Sync Removed

Main: Full nodes sync from both P2P (via go-header HeaderSyncService/DataSyncService) and DA. The executor broadcasts produced blocks to P2P peers. P2P events include DA height hints for targeted DA retrieval. The syncer runs a P2P worker loop alongside the DA follower.

This branch: All P2P sync is removed. Nodes sync exclusively from DA. No P2P broadcasting, no P2P stores, no P2P handler, no DA height hints.

Removed code

  • syncing/p2p_handler.go — entire file deleted
  • syncing/syncer_mock.go — P2P handler mock deleted
  • common/expected_interfaces.go — HeaderP2PBroadcaster/DataP2PBroadcaster types removed
  • P2P broadcasting in executing/executor.go removed
  • P2P worker loop in syncer removed
  • SourceP2P event source removed
  • DaHeightHints field removed from DAHeightEvent
  • headerStore/dataStore parameters removed from NewSyncer and component constructors
  • headerSyncService/dataSyncService parameters removed from aggregator component constructors
  • DAHintAppender interface removed from DA submitter
  • Separate SubmitHeaders/SubmitData replaced with single SubmitBlocks

Why

  • Removes network overhead from P2P gossip
  • Eliminates the complexity of two sync sources competing
  • DA is the single source of truth, reducing consistency issues
  • Removes libp2p dependency from the block package's hot path
  • Simplifies the syncer from 3 worker loops to 2 (process loop + pending worker)

4. DA Submitter Simplified

Main: DASubmitterAPI has two methods: SubmitHeaders and SubmitData, each with separate batching strategies, mutex locks, and retry loops.

This branch: DASubmitterAPI has a single SubmitBlocks method that takes headers and data together, creates combined blobs, signs them, and submits to a single namespace. One batching strategy, one mutex, one retry loop.

Why

  • Halves the submission loop complexity
  • Eliminates the envelope cache (no more retry-signing concern)
  • Single retry loop instead of two
  • Combined blobs submitted atomically — no partial header-without-data states

Migration Notes

  • Existing DA data from main branch is not readable by this branch (different blob format)
  • This branch requires a fresh start or a migration tool
  • The P2PSignedHeader and P2PData types still exist in types/ but are no longer used by the block package
  • External consumers of NewSyncComponents and NewAggregatorWithCatchupComponents must update their call sites