When migrating ‘real world’ VMs, we often see problems occurring as a result of the guest being aware it has been migrated. These problems seem to break down in to two categories - those which are kernel related (typically genuine bugs), and those which are userspace related (i.e. userspace applications that break after the migrate, even though the VM itself is still running). As an example of the second category, if you migrate a VM running kubernetes, then (depending on the exact networking model in use), you can end up in a state where the container networking is entirely broken once the migration has completed.
This session is intended to identify the problems seen, and discuss ideas as to how to resolve / workaround them - these might range from full guest unaware migration (e.g. we could discuss what might be required to achieve it), to more tactical changes (e.g. changes to avoid triggering udev events or similar during migration which are the likely cause of some userspace issues).