CPU and memory use of CI jobs
How much of their runner’s memory and CPU the GitHub Actions jobs that finished in the period used, measured on runners AutoPilot for CI launched and wherever else usage was captured; jobs on other runners count in unmeasured_runs. memory_used_ratio is peak memory over the runner’s memory and cpu_used_ratio average CPU over the runner’s cores; on a runner AutoPilot launched, the runner’s memory and cores are the size it launched, so a job that bursts past that size reads above 1; p50 and p90 are across measured runs. idle_memory_gb_hours sums (runner memory - peak) x run hours, and idle_cpu_core_hours sums (runner cores - average cores) x run hours, over measured runs. jobs gives one row per job and runner with at least one measured run, sorted by memory_used_ratio_max ascending unless sort and order say otherwise; memory_total_gb and cpu_total_cores are the size of the job’s latest run, and memory_used_ratio_max is memory_peak_gb_max, the highest peak across its runs, over that size; run_seconds_avg and run_seconds_max are the average and longest run time of the job’s measured runs. cpu gives the job’s CPU as a share of the runner’s cores across its measured runs: mean averages the runs’ average use, max is the highest peak, and p50 to p99 are the highest of the runs’ per-second percentiles; min, std_dev and last are not measured and stay 0. A job measured by the usage step or a runner AutoPilot launched shows as soon as its reading is stored; until GitHub’s job record arrives its run time is the length of the reading. series gives the measured runs and median ratios per UTC day. Defaults to the last 30 days; a period over 400 days is rejected.