Alertmanager should be cgroup aware

[uhthomas]

What did you do?

I recently changed the CPU in my server from an Intel 13600k with 20 threads to an EPYC 7763 with 128 threads. I run alertmanager in Kubernetes with a CPU limit of 100m.

What did you expect to see?

No change in performance.

What did you see instead? Under which circumstances?

Go is not cgroup aware. This means that Prometheus is being throttled in containerised environments with CPU limits.

The first half of the time series is with the 13600k, and the second half is with the EPYC 7763.

automaxprocs can use cgroups automatically.

Environment

• System information:

Linux 6.1.69-talos x86_64

• Alertmanager version:

alertmanager, version 0.26.0 (branch: main, revision: 7a3c189315da7b8ed2d5e05f122c4cf4873c1379)
  build user:       root@9178099891ea
  build date:       20240123-14:29:39
  go version:       go1.21.6
  platform:         linux/amd64
  tags:             netgo

• Prometheus version:

N/A

• Alertmanager configuration file:

N/A

• Prometheus configuration file:

N/A

• Logs:

N/A

[siavashs]

Fixed in #3837

[uhthomas]

Ironically, automaxprocs should probably be removed now that Go 1.25 does this natively.

[siavashs]

Alertmanager is not on Go 1.25 yet, we'll revisit this when we move to 1.25