Object Pooling

Bingsan uses sync.Pool from Go's standard library to reduce memory allocation pressure in hot paths.

Overview

Two types of pools are implemented:

Pool	Purpose	Default Size	Max Size
BufferPool	JSON serialization buffers	4 KB	64 KB
BytePool	OAuth token generation	32 bytes	32 bytes

How It Works

BufferPool

The BufferPool provides reusable bytes.Buffer instances for JSON serialization:

Request 1 ──► Get buffer ──► Serialize JSON ──► Return buffer ──► Pool
                  │                                  ▲
                  └──────────────────────────────────┘
                              Reused

Key characteristics:

Initial capacity: 4 KB (typical JSON metadata size)
Maximum size: 64 KB (oversized buffers are discarded)
Thread-safe via sync.Pool
Automatic reset on get

BytePool

Fixed size: 32 bytes for OAuth access token generation.

Usage Patterns

In API Handlers

func (h *Handler) GetTable(ctx *fiber.Ctx) error {
    buf := pool.GetBuffer()
    defer pool.PutBuffer(buf)  // Always return!

    encoder := json.NewEncoder(buf)
    if err := encoder.Encode(table); err != nil {
        return err
    }

    return ctx.Send(buf.Bytes())
}

Configuration

Constant	Value	Description
`DefaultBufferSize`	4096	Initial buffer capacity in bytes
`MaxBufferSize`	65536	Maximum buffer size before discard
`TokenSize`	32	Fixed size for token byte slices

Best Practices

Always Use `defer`

buf := pool.GetBuffer()
defer pool.PutBuffer(buf)  // Guaranteed return

Don't Hold References

// Wrong: Reference escapes
data := buf.Bytes()
pool.PutBuffer(buf)
return data  // data is now invalid!

// Correct: Copy if needed
data := make([]byte, buf.Len())
copy(data, buf.Bytes())
pool.PutBuffer(buf)
return data

Metrics

Pool performance is exposed via Prometheus:

Metric	Type	Description
`bingsan_pool_gets_total`	Counter	Total Get() operations
`bingsan_pool_returns_total`	Counter	Total Put() operations
`bingsan_pool_discards_total`	Counter	Oversized items discarded
`bingsan_pool_misses_total`	Counter	New allocations (pool empty)

Benchmarks

go test -bench=BenchmarkPool -benchmem ./tests/benchmark/...

Expected results:

Benchmark	Time	Allocs
BufferPool.Get/Put	~50ns	0
BufferPool.Concurrent	~100ns	0
BytePool.Get/Put	~30ns	0

Troubleshooting

High Discard Rate

If bingsan_pool_discards_total is increasing rapidly:

Cause: Many large responses exceeding 64KB
Impact: Reduced pool effectiveness
Solution: Consider increasing MaxBufferSize for schemas with 100+ columns

Object Pooling

On this page