Bug 1988252

Description of problem: The podman stack got updated recently in RHEL 8.5. This enabled the /stats API for unprivileged containers, possibly inadvertently.


Version-Release number of selected component (if applicable):

  conmon (2:2.0.27-1.module+el8.5.0+10387+8d85dbaf -> 2:2.0.29-1.module+el8.5.0+12014+438a5746)
  kernel (4.18.0-323.el8 -> 4.18.0-324.el8)
  podman (3.1.0-0.8.module+el8.5.0+10387+8d85dbaf -> 3.3.0-0.17.module+el8.5.0+12014+438a5746)
  runc (1.0.0-70.rc92.module+el8.5.0+10387+8d85dbaf -> 1.0.1-3.module+el8.5.0+12014+438a5746)

How reproducible: Always


Steps to Reproduce (as non-root):
1. Enable the API and start a container:

systemctl --user start podman.socket
podman run -it quay.io/libpod/busybox

2. Check usage stats on the CLI:

$ podman stats
Error: stats is not supported in rootless mode without cgroups v2
(this did not change)

3. Check usage stats on the API;

curl --unix-socket /run/user/1000/podman/podman.sock http://d/v1.24/libpod/containers/$(podman ps -q)/stats

Actual results:
With the previous versions this failed similarly to the CLI:

{"cause":"no support for CGroups V1 in rootless environments","message":"failed to obtain Container 283b58f272eb stats: unable to load cgroup at /user.slice/user-1000.slice/user/user.slice/podman-1565.scope/283b58f272ebd8ad68c647d6c50ac11ed871c30f17a0e61151c78c1d547d823a: no support for CGroups V1 in rootless environments","response":500}

with the current versions, this now starts to work, but reports a lot of wrong numbers:

{"read":"2021-07-30T02:59:55.76241298-04:00","preread":"2021-07-30T02:59:50.758203107-04:00","pids_stats":{"current":15},"blkio_stats":{"io_service_bytes_recursive":[],"io_serviced_recursive":null,"io_queue_recursive":null,"io_service_time_recursive":null,"io_wait_time_recursive":null,"io_merged_recursive":null,"io_time_recursive":null,"sectors_recursive":null},"num_procs":0,"storage_stats":{},"cpu_stats":{"cpu_usage":{"total_usage":0,"usage_in_kernelmode":0,"usage_in_usermode":0},"system_cpu_usage":8325461890,"cpu":0,"throttling_data":{"periods":0,"throttled_periods":0,"throttled_time":0}},"precpu_stats":{"cpu_usage":{"total_usage":0,"usage_in_kernelmode":0,"usage_in_usermode":0},"system_cpu_usage":8292281354,"cpu":0,"throttling_data":{"periods":0,"throttled_periods":0,"throttled_time":0}},"memory_stats":{"usage":48406528,"max_usage":9223372036854771712,"limit":9223372036854771712},"name":"festive_goldberg","Id":"7151de8987656745a6e937ce6b35d6cacd8200ae5e7c015b65f03e931138e9a0","networks":{"network":{"rx_bytes":0,"rx_packets":0,"rx_errors":0,"rx_dropped":0,"tx_bytes":0,"tx_packets":0,"tx_errors":0,"tx_dropped":0}}}


At least blkio_stats, num_procs, networks, and cpu_stats are always zero, and I suspect with cgroupsv1  there is not much chance of actually getting them. You can trigger some CPU usage with `dd if=/dev/urandom of=/dev/null` but the numbers stay at 0. Same with network traffic with `wget`.

Only memory_stats.usage seems correct , when I do `MEMBLOB=$(yes | dd bs=1M count=200 iflag=fullblock)` the usage does go up by ~ 300 MB.


Expected results: If stats reports numbers, they should be correct. This currently causes invalid numbers to be displayed in cockpit-podman.


Additional info: