Clone of go.uber.go/automaxprocs without the logging.
Automatically set GOMAXPROCS to match Linux container CPU quota.
go get -u github.com/overjetdental/automaxprocs
import _ "github.com/overjetdental/automaxprocs"
func main() {
// Your application logic here.
}Data measured from Uber's internal load balancer. We ran the load balancer with 200% CPU quota (i.e., 2 cores):
| GOMAXPROCS | RPS | P50 (ms) | P99.9 (ms) |
|---|---|---|---|
| 1 | 28,893.18 | 1.46 | 19.70 |
| 2 (equal to quota) | 44,715.07 | 0.84 | 26.38 |
| 3 | 44,212.93 | 0.66 | 30.07 |
| 4 | 41,071.15 | 0.57 | 42.94 |
| 8 | 33,111.69 | 0.43 | 64.32 |
| Default (24) | 22,191.40 | 0.45 | 76.19 |
When GOMAXPROCS is increased above the CPU quota, we see P50 decrease slightly, but see significant increases to P99. We also see that the total RPS handled also decreases.
When GOMAXPROCS is higher than the CPU quota allocated, we also saw significant throttling:
$ cat /sys/fs/cgroup/cpu,cpuacct/system.slice/[...]/cpu.stat
nr_periods 42227334
nr_throttled 131923
throttled_time 88613212216618
Once GOMAXPROCS was reduced to match the CPU quota, we saw no CPU throttling.
All APIs are finalized, and no breaking changes will be made in the 1.x series
of releases. Users of semver-aware dependency management systems should pin
automaxprocs to ^1.
We encourage and support an active, healthy community of contributors — including you! Details are in the contribution guide and the code of conduct. The automaxprocs maintainers keep an eye on issues and pull requests, but you can also report any negative conduct to [email protected]. That email list is a private, safe space; even the automaxprocs maintainers don't have access, so don't hesitate to hold us to a high standard.
Released under the MIT License.