Description of problem:
During an overcloud deploy and subsequent stack updates, an md5sum is taken of the Octavia amphora image file (/usr/share/rhosp-director-images/octavia-amphora.qcow2). This is compared to the checksum of the existing amphora image in glance before deciding whether or not to upload the amphora image file into glance.
These checksums are always different, resulting in the image being unnecessarily uploaded on every overcloud stack update. The reason is that the local Octavia image file is converted from qcow2 to raw format before being uploaded into glance. This occurs after the checksum is taken, not before.
Manually performing an md5sum on the converted image (in my case /tmp/ansible.mRsIIP/octavia-amphora.img) shows that the checksums match between the converted image and the glance image checksum of an existing image.
The playbook logic needs to be fixed, probably to perform the image conversion first, then run a checksum and compare to the glance image checksum.
Version-Release number of selected component (if applicable):
octavia-amphora-image-x86_64-13.0-20190304.1.el7ost.noarch
octavia-amphora-image-13.0-20190304.1.el7ost.noarch
puppet-octavia-12.4.0-8.el7ost.noarch
How reproducible:
Always
Steps to Reproduce:
1. Deploy OSP13 overcloud with Octavia
2. Run `openstack overcloud deploy` again
3. Check glance images
Actual results:
Octavia amphora image is continually re-uploaded
Expected results:
checksum should confirm that image upload is not necessary
Additional info:
Engineering acknowledges this issue. However, how does this negatively impact customers other than taking a few more seconds on stack update?
This BZ was set with severity by the reporter and received priority low by Engineering. Please let us know if there's a considerable impact we have not considered, we can reconsider priority.