hyperledger / fabric-x-committer / 21895916992

90%

main: 90%

Build Type

Pull #316

github

Committed by cendhu

Commit Message

Introduce UnmarshalEnvelopeLite to speed up mapBlock

  Replace full proto.Unmarshal of Envelope→Payload→ChannelHeader (3 chained
  deserializations) with direct protobuf wire-format scanning that extracts
  only the needed fields (HeaderType, TxID, Data). This reduces envelope
  parsing from 287 ns/op to 72 ns/op — a 4x speedup with 7 fewer allocations.

  BenchmarkUnwrapEnvelope-14        4079934    286.7 ns/op   616 B/op   9 allocs/op
  BenchmarkUnwrapEnvelopeLite-14   16929306     71.5 ns/op    72 B/op   2 allocs/op

  [New mapBlock]
  BenchmarkMapBlock/txs=500-14                   2850            400160 ns/op          663662 B/op      13579 allocs/op
  BenchmarkMapBlock/txs=500-14                   2599            405622 ns/op          664974 B/op      13594 allocs/op
  BenchmarkMapBlock/txs=500-14                   2767            408145 ns/op          657391 B/op      13505 allocs/op

  [Old mapBlock]
  BenchmarkMapBlock/txs=500-14                   1497            677487 ns/op         1165977 B/op      17648 allocs/op
  BenchmarkMapBlock/txs=500-14                   1584            659822 ns/op         1168212 B/op      17674 allocs/op
  BenchmarkMapBlock/txs=500-14                   1808            635695 ns/op         1158503 B/op      17561 allocs/op

  With UnmarshalEnvelopeLite, the new mapBlock is able to process 1.7x more blocks.

Signed-off-by: Senthilnathan <cendhu@gmail.com>

Pull Request Pull Request #316: Introduce UnmarshalEnvelopeLite to speed up mapBlock

Run Details

87 of 109 new or added lines in 2 files covered. (79.82%)

2 existing lines in 1 file now uncovered.

9131 of 10291 relevant lines covered (88.73%)

1.29 hits per line