We are also seeing quite common, but probably not the same issue with Nova, with slightly different stack trace, but same result: https://paste.openstack.org/show/b3sOW4WSg74s3qk5vCgB/
It also can be the same issue, just logs from compute rather then API side.
Though, from what we have tested - heartbeat_in_pthread: False for Nova API seems to improve things with this one.
We're still monitoring state and it's probably early to make final conclusions. But I kinda do see how that can be related.
As eventlet import in oslo.messaging happens only for enabled heartbeat_in_pthread, and was introduced with this option: https://github.com/openstack/oslo.messaging/commit/22f240b82fffbd62be8568a7d0d3369134596ace
So disabling usage of eventlet seems to help in our case.
We are also seeing quite common, but probably not the same issue with Nova, with slightly different stack trace, but same result: /paste. openstack. org/show/ b3sOW4WSg74s3qk 5vCgB/
https:/
It also can be the same issue, just logs from compute rather then API side.
Though, from what we have tested - heartbeat_ in_pthread: False for Nova API seems to improve things with this one.
We're still monitoring state and it's probably early to make final conclusions. But I kinda do see how that can be related.
As eventlet import in oslo.messaging happens only for enabled heartbeat_ in_pthread, and was introduced with this option: /github. com/openstack/ oslo.messaging/ commit/ 22f240b82fffbd6 2be8568a7d0d336 9134596ace
https:/
So disabling usage of eventlet seems to help in our case.