Performance Optimization for minVV and maxVV calculation

[chacha912]

What would you like to be added:

Performance analysis of VV benchmark tests(#1150 ) revealed significant time and memory consumption in Version Vector Min/Max operations. This needs optimization to improve overall performance.

Current Performance Analysis

cpu-v057.prof	mem-v057.prof

• CPU Profile
  • UpdateAndFindMinSyncedVersionVector: 20.48s (33.53%)
  • ApplyChanges: 9.83s (16.09%)
  • runtime.gcMarkWorker: 13.30s (21.81%)
• Memory Profile
  • UpdateAndFindMinSyncedVersionVector > VersionVector Min: 29,524.84MB (58.10%)
  • ApplyChangePack > VersionVector Max: 14,841.02MB (29.20%)

1. MinVV Calculation (PushPull Operation)

• Current implementation calculates minVV incrementally, using lots of memory

• yorkie/server/backend/database/mongo/client.go

  Lines 1265 to 1282 in 2fa676d

  	for _, vvi := range versionVectorInfos {
  	if clientInfo.ID == vvi.ClientID {
  	continue
  	}
  	if minVersionVector == nil {
  	minVersionVector = vvi.VersionVector
  	continue
  	}
  	minVersionVector = minVersionVector.Min(vvi.VersionVector)
  	}
  	if minVersionVector == nil {
  	minVersionVector = versionVector
  	}
  	// 02-2. Compute min version vector with current client's version vector.
  	minVersionVector = minVersionVector.Min(versionVector)

2. MaxVV Calculation (Attach Operation)

• Used during SyncClocks when applying remote changes
• Performance degrades with:
  • Increasing number of clients
  • Accumulating changes
• Note: Current benchmark tests disable snapshots during attach

Steps to Reproduce

1. Run benchmark with profiling:

go test -tags bench -benchmem -bench=BenchmarkVersionVector ./test/bench -memprofile=mem.prof -cpuprofile=cpu.prof

2. Analyze profile:

# View single profile
go tool pprof [file path]
(pprof) web  # Requires graphviz package

# Compare two profiles
go tool pprof -base mem-v057.prof mem.prof

• Profile data is available in the attached compressed files for reference. Test conducted with 1000 clients. prof.zip

Why is this needed:

To enhance performance during large-scale concurrent editing with multiple clients