It would be useful if node_exporter could expose the current CPU Energy Performance Preference (EPP), similar to node_cpu_scaling_governor.
In practical terms, the governor defines the general CPU frequency/performance scaling policy, while EPP is a finer hint to the CPU about how strongly it should prefer performance versus energy efficiency within that policy. For example, with the powersave governor active, EPP can still range from performance to balance_performance, balance_power, or power, which can noticeably change responsiveness and power consumption.
Linux already exposes this per CPU through:
/sys/devices/system/cpu/cpu*/cpufreq/energy_performance_preference
/sys/devices/system/cpu/cpu*/cpufreq/energy_performance_available_preferences
Something like this would make EPP changes easy to monitor historically:
node_cpu_energy_performance_preference{cpu="0",preference="balance_power"} 1
node_cpu_energy_performance_preference{cpu="0",preference="performance"} 0
Since governor state is already exposed this way, EPP seems like a natural addition to the cpufreq collector.
Thank you. :)
It would be useful if node_exporter could expose the current CPU Energy Performance Preference (EPP), similar to
node_cpu_scaling_governor.In practical terms, the governor defines the general CPU frequency/performance scaling policy, while EPP is a finer hint to the CPU about how strongly it should prefer performance versus energy efficiency within that policy. For example, with the
powersavegovernor active, EPP can still range fromperformancetobalance_performance,balance_power, orpower, which can noticeably change responsiveness and power consumption.Linux already exposes this per CPU through:
Something like this would make EPP changes easy to monitor historically:
Since governor state is already exposed this way, EPP seems like a natural addition to the cpufreq collector.
Thank you. :)