Skip to content

BENCHMARKS.md

Latest commit

 

History

History
61 lines (41 loc) · 1.83 KB

BENCHMARKS.md

File metadata and controls

61 lines (41 loc) · 1.83 KB
Raw

Benchmark plan

Benchmarks should be honest and not overclaim.

Run:

go test -bench=. -benchmem ./...

For more stable local numbers, run multiple samples:

go test -run='^$' -bench=. -benchmem -count=5 ./...

Benchmark cases

  1. Plain channel send/recv same goroutine.
  2. chanprobe.Queue Send/Recv same goroutine.
  3. Plain channel producer/consumer.
  4. chanprobe.Queue producer/consumer.
  5. TrySend/TryRecv.
  6. Snapshot under load.

Interpreting results

Compare related pairs:

  • BenchmarkChannelSameGoroutine vs BenchmarkQueueSameGoroutine
  • BenchmarkChannelProducerConsumer vs BenchmarkQueueProducerConsumer

Native channels are the baseline. chanprobe.Queue is expected to cost more because it tracks counters, wait durations, item age, close state, and supports context-aware blocking operations.

BenchmarkQueueTrySendTryRecv measures the non-blocking fast path.

BenchmarkQueueSnapshot measures the cost of reading observability data from a populated queue.

Example local result on an AMD Ryzen 3 4300U:

BenchmarkChannelSameGoroutine-4       30.80 ns/op    0 B/op  0 allocs/op
BenchmarkQueueSameGoroutine-4       2293 ns/op       0 B/op  0 allocs/op
BenchmarkChannelProducerConsumer-4    51.11 ns/op    0 B/op  0 allocs/op
BenchmarkQueueProducerConsumer-4    2389 ns/op       0 B/op  0 allocs/op
BenchmarkQueueTrySendTryRecv-4      2262 ns/op       0 B/op  0 allocs/op
BenchmarkQueueSnapshot-4            1164 ns/op       0 B/op  0 allocs/op

Treat these as local reference numbers, not a universal performance claim.

README benchmark wording

Do not claim chanprobe is faster than channels.

Expected positioning:

chanprobe adds observability and context-aware queue operations. It has overhead compared to native channels. Use it at important async boundaries where visibility matters.