v0.3.5

• batchSize

v0.3.4

• increase batchsize

use variant url

v0.3.3

v0.3.2 - Adaptive Batching

Performance and Reliability Improvements

Adaptive Batching Strategy

Replaced single large fetch with intelligent batching approach:

• Batch size: 5,000 messages per batch
• Adaptive timeout: Scales with batch size (batch_size/20 ms, minimum 100ms)
• Smart termination: Stops when server returns ErrNoMessages

Performance Results

Small streams (15k messages):

• v0.3.0/v0.3.1: 1-2s
• v0.3.2: ~750ms → 2-3x faster ⚡

Medium streams (65k messages):

• v0.3.0/v0.3.1: 1-2s (inconsistent)
• v0.3.2: ~3.3s → Consistent and reliable ✅

Large streams (200k messages):

• v0.3.0/v0.3.1: Timeouts or incomplete results
• v0.3.2: ~10s → Handles correctly ✅

Fixes

• ❌ v0.3.1 bug fixed: invalid argument: batch size must be at least 1 error when NumPending was 0
• ✅ Race condition eliminated: No longer uses NumPending approach that caused inconsistent results
• ✅ Consistency guaranteed: Timeout per batch ensures server completes gathering
• ✅ Scales correctly: Time proportional to actual data size

Dependencies

Updated all dependencies to latest versions:

• AWS SDK v2: v1.32.5 → v1.40.1
• imroc/req: v3.48.0 → v3.56.0
• Ginkgo: v2.20.2 → v2.27.2
• And more...

Migration from v0.3.1

If you're on v0.3.1: Upgrade immediately - it has a critical bug causing batch size must be at least 1 errors

Recommended version: v0.3.2 (this release) for both performance and reliability

v0.3.1 - Performance Optimization

Performance Improvements

LastPerSubject Optimization

• 19.42x faster execution using NumPending approach
• Eliminates arbitrary timeout waits by checking exact message count
• Uses ConsumerInfo NumPending + FetchNoWait for optimal performance
• Maintains 100% consistency across all runs

Technical Changes

• Reuse context instead of creating multiple context.Background instances
• Reduced InactiveThreshold from 5s to 1s for faster ephemeral consumer cleanup
• Removed old timeout-based implementation

Benchmark Results

• Old method: ~1s per call (timeout-based)
• New method: ~50ms per call (NumPending-based)
• 5/5 runs consistently return identical results

v0.3.0 - NATS Upgrade & Performance Improvements

🚀 Features

• Upgrade nats.go to v1.47.0 (from v1.37.0) for latest performance improvements
• Optimize for consistent results with

⚡ Performance Improvements

• Fix inconsistency by using with timeout instead of
• Use ephemeral consumer with for automatic cleanup
• Reduce timeout from 30s to 1s for faster operation
• Remove unnecessary consumer config (, )
• Add consistency testing with 5 runs to verify reliable results

📦 Dependencies Updated

• nats.go: v1.37.0 → v1.47.0
• nkeys: v0.4.7 → v0.4.12
• klauspost/compress: v1.17.9 → v1.18.2
• golang.org/x/crypto: v0.29.0 → v0.45.0
• golang.org/x/net: v0.31.0 → v0.47.0
• golang.org/x/sys: v0.27.0 → v0.38.0

✅ Backward Compatibility

No breaking changes - all APIs maintain backward compatibility. Existing code will work without modifications.

🔗 Full Changelog

v0.2.9...v0.3.0

Upload video minus debug code

v0.2.33

Video uploads via cloudflare batch upload

v0.2.32

Media uploads via cloudflare batch upload

v0.2.31

v0.2.30

Changes

• Remove logrus package and its components (5147813)