ceilometer_memory_usage fails to report metrics in case where VMs disks are stored at ceph I have an OSPdO running OSP 16.2 in HCI env. I have noticed that there is no data in VM Memory Utilization panel in Grafana. After checking compute.log in ceilometer directory on compute nodes I see the following message ceilometer_memory_usage shows no data 6:56 2023-06-28 03:33:36.063 15 WARNING ceilometer.compute.pollsters [-] memory.usage statistic in not available for instance 93366f22-da75-4df1-bb64-ef5710d123a7: ceilometer.compute.pollsters.NoVolumeException
Looks like libvirt itself cannot expose the memory statistics of an instance in an HCI env ~~~ [root@computehci-1 /]# virsh list Id Name State ----------------------------------- 1 instance-00000001 running 2 instance-00000003 running [root@computehci-1 /]# virsh dommemstat 1 actual 524288 last_update 0 rss 228372 [root@computehci-1 /]# virsh dommemstat 2 actual 524288 last_update 0 rss 209964 ~~~ From a compute node of non-HCI env but nova uses ceph as a backend ~~~ [root@compute-0 /]# virsh list Id Name State ----------------------------------- 1 instance-00000001 running [root@compute-0 /]# virsh dommemstat instance-00000001 actual 524288 swap_in 0 swap_out 0 major_fault 0 minor_fault 84757 unused 402840 available 477392 usable 431064 last_update 1688384922 disk_caches 26528 hugetlb_pgalloc 0 hugetlb_pgfail 0 rss 353636 ~~~ Ceilometer uses these readings to generate memory usage samples [1] which are missing in HCI env. We must analyze this behaviour in OSP17 as well. [1] https://github.com/openstack/ceilometer/blob/stable/train/ceilometer/compute/virt/libvirt/inspector.py#L171-L185