Hello, I have a volume-backed instance with 16 vCPU and 64GB RAM. The instance uses 60GB RAM out of 64GB (used: 22GB; buff/cache 38GB). When I do a live migration of this instance, it fails without any timeouts. It copies almost all the RAM (within 150 - 250 seconds) to the target compute host without any problems according to the logs. Then the instance is paused to copy the rest of the RAM. Everything seems to be working correctly up to this point, but then the instance resumes and the following error message appears:
Live Migration failure: operation failed: migration out job: unexpectedly failed: libvirt.libvirtError: operation failed: migration out job: unexpectedly failed
Unfortunately, this error message does not say much. It doesn't look like it's due to any timeouts or short downtimes, but I still tested different (higher) values for the following configurations. Unfortunately without success. - live_migration_completion_timeout - live_migration_timeout_action: abort / force_complete (pause) - live_migration_downtime - live_migration_downtime_steps - live_migration_downtime_delay - live_migration_permit_auto_converge: True / False
All other instances on the same source and destination hosts can be live migrated without any issues. This instance can also be successfully live migrated after a restart, as it is probably not yet heavily loaded. After a few hours, however, the live migration no longer works.
Any ideas what the problem could be?
Logs: - nova-compute.log from source compute host: https://paste.openstack.org/show/bJMFxnPKQBEVaPakud61/ - i found this Traceback using journalctrl: https://paste.openstack.org/show/bIS4GFAd2RJ5fHVN9I8d/ - there was also an error in /var/log/libvirt/qemu/: https://paste.openstack.org/show/bImT89IelDcXXBPSgTCO/
Enviroment: - Libvirt: 8.0.0 - QEMU: 4.2.1 - Nova: 25.1.1 - OpenStack: Yoga - Compute operating system: Ubuntu 20.04