Ask Your Question

Revision history [back]

I don't find the term "rate" on that page, but I understand IOPS as the IOs per second, both read and write, whereas read requests metrics cover read requests, and write request metrics cover write requests only.

I don't find the term "rate" on that page, but I understand IOPS as the IOs per second, both read and write, whereas read requests metrics cover read requests, and write request metrics cover write requests only.


Regarding source of measurement information, first see https://docs.openstack.org/ceilometer/latest/admin/telemetry-data-collection.htm. The first method is services providing information to Ceilometer via the message queue. After all, services should know everything about their objects; e.g. Nova should know all details about its instances. This method is named notification.

Since the services don't contribute all the samples that Ceilometer would like to have, it also has its own polling agents that get information from sources like hypervisors. An instance's disk I/O is most likely sampled by a pollster running on the compute host. Such a pollster knows how to use libvirt or whatever other hypervisor API is suitable to gather the required data. See also the lists on https://docs.openstack.org/ceilometer/latest/admin/telemetry-measurements.html.