Default to cgroup (sysfs) stat collection on native Linux hosts, falling back to the Docker API only on linuxkit or read failure.
Note: reported container memory usage may step down on hosts previously using stats-type: docker, because sysfs excludes inactive file cache (matching docker stats).
Block-I/O readings may also shift on cgroup v1 hosts, because the sysfs path reads blkio.throttle.io_service_bytes while the Docker API path reads blkio_stats.io_service_bytes_recursive (which sums across nested cgroups).
Dashboards or autoscaling thresholds tuned to the old values should be re-evaluated after upgrade.