Comment 6 for bug 1773945

I hit something similar last week in python-openstackclient when listing servers, I was getting a marker not found error even though I wasn't passing a marker, but something in the CLI (or nova python API binding code) was. I ran into this after rabbitmq crashed during scheduling a large number of instances in a single request, and I think I also noticed that there were instance_mappings records with no cell_id set.

I believe the reason the CLI was passing a marker to the server, even though I wasn't specifying a marker (or a limit) was because of this code in python-novaclient:

https://github.com/openstack/python-novaclient/blob/10.2.0/novaclient/v2/servers.py#L863

That logic says, on the first pass to the server, don't pass a marker or a limit, but if you get results back, keep hitting the server and this time use a marker from the last result set, and that's where I was getting the marker not found on my corrupted instances.

The default paging behavior with 'nova list' was changed recently which might have introduced this regression:

https://review.openstack.org/#/c/534222/