Runner jobs now report real machine metrics (CPU, memory, network throughput, disk) sampled from inside the VM itself, so the Metrics tab in the job view is populated for both macOS and Linux runners. Previously metrics were sampled from the controller but produced nothing, leaving the tab empty.
Sampling runs for the full duration of every job and is best-effort: a failed sample or POST is skipped and never blocks the job.
Note: metrics will appear once the new runner images are rolled out (older warm Pods on prior images won’t post until recycled onto the new image). On macOS, sampling stops if a single job runs past the SA token’s 1 hour TTL.