search

appsignal / probes-rs

Rust library to read out system stats from a machine running Unix
MIT License
15 stars 2 forks source link

Normalize container CPU percentages for cgroups v1 #77

Closed tombruijn closed 9 months ago

tombruijn commented 10 months ago

Follow up of #67.

Support CPU uage percentage normalization for cgroups v1.

Not sure how to determine the number of CPUs yet.

Possibly the /sys/fs/cgroup/cpu/cpu.cfs_quota_us file is useful. For one CPU it returns 100000 and for two CPUs it returns 200000.

cpu.cfs_quota_us: The maximum time in microseconds during each cfs_period_us in for the current group will be allowed to run. For instance, if it is set to half of cpu_period_us, the cgroup will only be able to peak run for 50 % of the time. One should note that this represents aggregate time over all CPUs in the system. Therefore, in order to allow full usage of two CPUs, for instance, one should set this value to twice the value of cfs_period_us.

cpu.cfs_period_us: The duration in microseconds of each scheduler period, for bandwidth decisions. This defaults to 100000us or 100ms. Larger periods will improve throughput at the expense of latency, since the scheduler will be able to sustain a cpu-bound workload for longer. The opposite of true for smaller periods. Note that this only affects non-RT tasks that are scheduled by the CFS scheduler.

tombruijn commented 10 months ago

Requested in https://app.intercom.com/a/inbox/yzor8gyw/inbox/shared/unassigned/conversation/16410700277823?view=List