Public bug reported: When live migrating libvirt instances, the destination host doesn't properly check the available disk space when using image files and doing overcommit, leading to migration failures.
Trace: http://paste.openstack.org/raw/798895/ It seems to be using resource tracker information that is not aware of disk overcommitment, so we end up with negative values. The "local_gb_used" value reflects the total allocated space, not the actually used disk space. https://github.com/openstack/nova/blob/20.4.0/nova/compute/resource_tracker.py#L1254 The same incorrect values will be reported by "openstack hypervisor show": http://paste.openstack.org/raw/798898/ Additionally, the "disk_over_commit" boolean flag is incorrectly checked. The driver checks if the field exists as part of the "dest_check_data" dict but doesn't actually check its value. https://github.com/openstack/nova/blob/20.4.0/nova/virt/libvirt/driver.py#L8224 The "disk_over_commit" parameter is deprecated. Recent Nova API versions do not use it, which bypasses the disk allocation check on the libvirt driver side. This might be used as a workaround (e.g. using nova client instead of the openstack client or horizon), but this is not ideal. ** Affects: nova Importance: Undecided Status: New ** Tags: libvirt live-migration resource-tracker -- You received this bug notification because you are a member of Yahoo! Engineering Team, which is subscribed to OpenStack Compute (nova). https://bugs.launchpad.net/bugs/1899139 Title: Live migrations don't properly handle disk overcommitment Status in OpenStack Compute (nova): New Bug description: When live migrating libvirt instances, the destination host doesn't properly check the available disk space when using image files and doing overcommit, leading to migration failures. Trace: http://paste.openstack.org/raw/798895/ It seems to be using resource tracker information that is not aware of disk overcommitment, so we end up with negative values. The "local_gb_used" value reflects the total allocated space, not the actually used disk space. https://github.com/openstack/nova/blob/20.4.0/nova/compute/resource_tracker.py#L1254 The same incorrect values will be reported by "openstack hypervisor show": http://paste.openstack.org/raw/798898/ Additionally, the "disk_over_commit" boolean flag is incorrectly checked. The driver checks if the field exists as part of the "dest_check_data" dict but doesn't actually check its value. https://github.com/openstack/nova/blob/20.4.0/nova/virt/libvirt/driver.py#L8224 The "disk_over_commit" parameter is deprecated. Recent Nova API versions do not use it, which bypasses the disk allocation check on the libvirt driver side. This might be used as a workaround (e.g. using nova client instead of the openstack client or horizon), but this is not ideal. To manage notifications about this bug go to: https://bugs.launchpad.net/nova/+bug/1899139/+subscriptions -- Mailing list: https://launchpad.net/~yahoo-eng-team Post to : yahoo-eng-team@lists.launchpad.net Unsubscribe : https://launchpad.net/~yahoo-eng-team More help : https://help.launchpad.net/ListHelp
