A more precise sleep for Go.
time.Sleep only promises to sleep for at least the duration specified.
The amount of overshooting can be quite large and totally random.
When microsecond precision is needed, powernap can be the solution you need.
Depending on your system, it becomes relatively precise above 1-10 µs.
powernap was created for the cases, where you're willing to give up CPU cycles for precision.
When possible, a safe part of the duration is done with time.Sleep, which probably happens around 1-10 ms, the rest is done by looping until the total duration has elapsed.
powernap can undershoot and makes no promises of sleeping "at least" for the duration given.
The promise is: Spend CPU cycles to get as close as possible to the wanted duration, but be smart about it.
It can be hard to describe the precise effect, so the following benchmark will show you the differences.
You can run this benchmark yourself with go test -bench .
The last value ns/op should be as close as possible to the test names, e.g.: 100ns.
Concurrently on 8 cores:
$ go test -run none -bench Sleep
goos: darwin
goarch: amd64
pkg: github.com/morphar/powernap
cpu: Intel(R) Core(TM) i7-1068NG7 CPU @ 2.30GHz
BenchmarkNativeSleep/1ns-8 3668739 324.8 ns/op
BenchmarkNativeSleep/10ns-8 3734082 320.1 ns/op
BenchmarkNativeSleep/100ns-8 3241970 370.3 ns/op
BenchmarkNativeSleep/1µs-8 256581 4750 ns/op
BenchmarkNativeSleep/10µs-8 68962 17406 ns/op
BenchmarkNativeSleep/100µs-8 10000 120359 ns/op
BenchmarkNativeSleep/1ms-8 972 1263703 ns/op
BenchmarkNativeSleep/10ms-8 100 10777938 ns/op
BenchmarkNativeSleep/100ms-8 10 101013503 ns/op
BenchmarkNativeSleep/1s-8 1 1001136413 ns/op
BenchmarkSleep/1ns-8 304529434 3.826 ns/op
BenchmarkSleep/10ns-8 312554546 3.814 ns/op
BenchmarkSleep/100ns-8 313016788 3.922 ns/op
BenchmarkSleep/1µs-8 1326577 915.8 ns/op
BenchmarkSleep/10µs-8 119400 10098 ns/op
BenchmarkSleep/100µs-8 10000 100199 ns/op
BenchmarkSleep/1ms-8 1198 1000308 ns/op
BenchmarkSleep/10ms-8 100 10000445 ns/op
BenchmarkSleep/100ms-8 10 100008614 ns/op
BenchmarkSleep/1s-8 1 1000001382 ns/op
BenchmarkSleepTight/1ns-8 305970579 3.951 ns/op
BenchmarkSleepTight/10ns-8 311431710 3.840 ns/op
BenchmarkSleepTight/100ns-8 311112458 3.911 ns/op
BenchmarkSleepTight/1µs-8 1328527 905.6 ns/op
BenchmarkSleepTight/10µs-8 120620 10033 ns/op
BenchmarkSleepTight/100µs-8 10000 100705 ns/op
BenchmarkSleepTight/1ms-8 1200 1000417 ns/op
BenchmarkSleepTight/10ms-8 100 10000474 ns/op
BenchmarkSleepTight/100ms-8 10 100003390 ns/op
BenchmarkSleepTight/1s-8 1 1000001256 ns/opSingle core:
dan@DansMac powernap % go test -run none -bench Sleep -cpu 1
goos: darwin
goarch: amd64
pkg: github.com/morphar/powernap
cpu: Intel(R) Core(TM) i7-1068NG7 CPU @ 2.30GHz
BenchmarkNativeSleep/1ns 5196897 246.4 ns/op
BenchmarkNativeSleep/10ns 5188018 234.8 ns/op
BenchmarkNativeSleep/100ns 4249675 271.0 ns/op
BenchmarkNativeSleep/1µs 179382 9319 ns/op
BenchmarkNativeSleep/10µs 73255 17105 ns/op
BenchmarkNativeSleep/100µs 10000 121338 ns/op
BenchmarkNativeSleep/1ms 1053 1189934 ns/op
BenchmarkNativeSleep/10ms 100 10744706 ns/op
BenchmarkNativeSleep/100ms 10 100730162 ns/op
BenchmarkNativeSleep/1s 1 1001129364 ns/op
BenchmarkSleep/1ns 306946417 3.837 ns/op
BenchmarkSleep/10ns 306756348 3.830 ns/op
BenchmarkSleep/100ns 309193824 3.897 ns/op
BenchmarkSleep/1µs 1242044 950.1 ns/op
BenchmarkSleep/10µs 119083 10107 ns/op
BenchmarkSleep/100µs 10000 100318 ns/op
BenchmarkSleep/1ms 1198 1000572 ns/op
BenchmarkSleep/10ms 100 10001685 ns/op
BenchmarkSleep/100ms 10 100004715 ns/op
BenchmarkSleep/1s 1 1000001456 ns/op
BenchmarkSleepTight/1ns 299916702 3.885 ns/op
BenchmarkSleepTight/10ns 308808326 3.813 ns/op
BenchmarkSleepTight/100ns 307965261 3.917 ns/op
BenchmarkSleepTight/1µs 1268040 950.3 ns/op
BenchmarkSleepTight/10µs 119745 10036 ns/op
BenchmarkSleepTight/100µs 10000 100139 ns/op
BenchmarkSleepTight/1ms 1197 1000479 ns/op
BenchmarkSleepTight/10ms 100 10000683 ns/op
BenchmarkSleepTight/100ms 10 100000765 ns/op
BenchmarkSleepTight/1s 1 1000001293 ns/opThe basic idea came from the Rust crate (library) spin_sleep.