One problem that is recurrent on our deployments is the image vs. nova-compute backend issues.
Nova-compute can accept following formats as backend: raw, qcow2 or rbd (which I believe is raw as well).
On the other side, when we upload a new image to Glance, we can upload a raw or qcow2 image.
Whenever we get a mismatch, cinder-volume process will run the conversion between the image we have and what it should be deployed.
That takes a lot of IOPS and processing for cinder-volume and may exhaust resources to other units that share the same disk as this cinder-volume.
Even if we come up with strategies to control cinder-volume, the ideal situation is not to have this mismatch in the first place.
Therefore, we should have a periodic check (or more than one) that raises alarms whenever a mismatch is detected.
That way, cloud operators can action and correct the image format before too much problems arise.
Hi,
One problem that is recurrent on our deployments is the image vs. nova-compute backend issues.
Nova-compute can accept following formats as backend: raw, qcow2 or rbd (which I believe is raw as well). On the other side, when we upload a new image to Glance, we can upload a raw or qcow2 image.
Whenever we get a mismatch, cinder-volume process will run the conversion between the image we have and what it should be deployed. That takes a lot of IOPS and processing for cinder-volume and may exhaust resources to other units that share the same disk as this cinder-volume.
Even if we come up with strategies to control cinder-volume, the ideal situation is not to have this mismatch in the first place.
Therefore, we should have a periodic check (or more than one) that raises alarms whenever a mismatch is detected. That way, cloud operators can action and correct the image format before too much problems arise.
Imported from Launchpad using lp2gh.
date created: 2021-03-15T08:45:33Z
owner: pguimaraes
assignee: None
the launchpad url