Orphaned resource allocations

Problem

There are orphaned resource allocations in the placement service which can cause resource providers to:

  • Appear to the scheduler to be more utilized than they really are

  • Prevent deletion of compute services

One scenario in which this could happen is a compute service host is having problems so the administrator forces it down and evacuates servers from it. Note that in this case “evacuates” refers to the server evacuate action, not live migrating all servers from the running compute service. Assume the compute host is down and fenced.

In this case, the servers have allocations tracked in placement against both the down source compute node and their current destination compute host. For example, here is a server vm1 which has been evacuated from node devstack1 to node devstack2:

$ openstack --os-compute-api-version 2.53 compute service list --service nova-compute
+--------------------------------------+--------------+-----------+------+---------+-------+----------------------------+
| ID                                   | Binary       | Host      | Zone | Status  | State | Updated At                 |
+--------------------------------------+--------------+-----------+------+---------+-------+----------------------------+
| e3c18c2d-9488-4863-b728-f3f292ec5da8 | nova-compute | devstack1 | nova | enabled | down  | 2019-10-25T20:13:51.000000 |
| 50a20add-cc49-46bd-af96-9bb4e9247398 | nova-compute | devstack2 | nova | enabled | up    | 2019-10-25T20:13:52.000000 |
| b92afb2e-cd00-4074-803e-fff9aa379c2f | nova-compute | devstack3 | nova | enabled | up    | 2019-10-25T20:13:53.000000 |
+--------------------------------------+--------------+-----------+------+---------+-------+----------------------------+
$ vm1=$(openstack server show vm1 -f value -c id)
$ openstack server show $vm1 -f value -c OS-EXT-SRV-ATTR:host
devstack2

The server now has allocations against both devstack1 and devstack2 resource providers in the placement service:

$ devstack1=$(openstack resource provider list --name devstack1 -f value -c uuid)
$ devstack2=$(openstack resource provider list --name devstack2 -f value -c uuid)
$ openstack resource provider show --allocations $devstack1
+-------------+-----------------------------------------------------------------------------------------------------------+
| Field       | Value                                                                                                     |
+-------------+-----------------------------------------------------------------------------------------------------------+
| uuid        | 9546fce4-9fb5-4b35-b277-72ff125ad787                                                                      |
| name        | devstack1                                                                                                 |
| generation  | 6                                                                                                         |
| allocations | {u'a1e6e0b2-9028-4166-b79b-c177ff70fbb7': {u'resources': {u'VCPU': 1, u'MEMORY_MB': 512, u'DISK_GB': 1}}} |
+-------------+-----------------------------------------------------------------------------------------------------------+
$ openstack resource provider show --allocations $devstack2
+-------------+-----------------------------------------------------------------------------------------------------------+
| Field       | Value                                                                                                     |
+-------------+-----------------------------------------------------------------------------------------------------------+
| uuid        | 52d0182d-d466-4210-8f0d-29466bb54feb                                                                      |
| name        | devstack2                                                                                                 |
| generation  | 3                                                                                                         |
| allocations | {u'a1e6e0b2-9028-4166-b79b-c177ff70fbb7': {u'resources': {u'VCPU': 1, u'MEMORY_MB': 512, u'DISK_GB': 1}}} |
+-------------+-----------------------------------------------------------------------------------------------------------+
$ openstack --os-placement-api-version 1.12 resource provider allocation show $vm1
+--------------------------------------+------------+------------------------------------------------+----------------------------------+----------------------------------+
| resource_provider                    | generation | resources                                      | project_id                       | user_id                          |
+--------------------------------------+------------+------------------------------------------------+----------------------------------+----------------------------------+
| 9546fce4-9fb5-4b35-b277-72ff125ad787 |          6 | {u'VCPU': 1, u'MEMORY_MB': 512, u'DISK_GB': 1} | 2f3bffc5db2b47deb40808a4ed2d7c7a | 2206168427c54d92ae2b2572bb0da9af |
| 52d0182d-d466-4210-8f0d-29466bb54feb |          3 | {u'VCPU': 1, u'MEMORY_MB': 512, u'DISK_GB': 1} | 2f3bffc5db2b47deb40808a4ed2d7c7a | 2206168427c54d92ae2b2572bb0da9af |
+--------------------------------------+------------+------------------------------------------------+----------------------------------+----------------------------------+

One way to find all servers that were evacuated from devstack1 is:

$ nova migration-list --source-compute devstack1 --migration-type evacuation
+----+--------------------------------------+-------------+-----------+----------------+--------------+-------------+--------+--------------------------------------+------------+------------+----------------------------+----------------------------+------------+
| Id | UUID                                 | Source Node | Dest Node | Source Compute | Dest Compute | Dest Host   | Status | Instance UUID                        | Old Flavor | New Flavor | Created At                 | Updated At                 | Type       |
+----+--------------------------------------+-------------+-----------+----------------+--------------+-------------+--------+--------------------------------------+------------+------------+----------------------------+----------------------------+------------+
| 1  | 8a823ba3-e2e9-4f17-bac5-88ceea496b99 | devstack1   | devstack2 | devstack1      | devstack2    | 192.168.0.1 | done   | a1e6e0b2-9028-4166-b79b-c177ff70fbb7 | None       | None       | 2019-10-25T17:46:35.000000 | 2019-10-25T17:46:37.000000 | evacuation |
+----+--------------------------------------+-------------+-----------+----------------+--------------+-------------+--------+--------------------------------------+------------+------------+----------------------------+----------------------------+------------+

Trying to delete the resource provider for devstack1 will fail while there are allocations against it:

$ openstack resource provider delete $devstack1
Unable to delete resource provider 9546fce4-9fb5-4b35-b277-72ff125ad787: Resource provider has allocations. (HTTP 409)

Solution

Using the example resources above, remove the allocation for server vm1 from the devstack1 resource provider. If you have osc-placement 1.8.0 or newer, you can use the openstack resource provider allocation unset command to remove the allocations for consumer vm1 from resource provider devstack1:

$ openstack --os-placement-api-version 1.12 resource provider allocation \
    unset --provider $devstack1 $vm1
+--------------------------------------+------------+------------------------------------------------+----------------------------------+----------------------------------+
| resource_provider                    | generation | resources                                      | project_id                       | user_id                          |
+--------------------------------------+------------+------------------------------------------------+----------------------------------+----------------------------------+
| 52d0182d-d466-4210-8f0d-29466bb54feb |          4 | {u'VCPU': 1, u'MEMORY_MB': 512, u'DISK_GB': 1} | 2f3bffc5db2b47deb40808a4ed2d7c7a | 2206168427c54d92ae2b2572bb0da9af |
+--------------------------------------+------------+------------------------------------------------+----------------------------------+----------------------------------+

If you have osc-placement 1.7.x or older, the unset command is not available and you must instead overwrite the allocations. Note that we do not use openstack resource provider allocation delete here because that will remove the allocations for the server from all resource providers, including devstack2 where it is now running; instead, we use openstack resource provider allocation set to overwrite the allocations and only retain the devstack2 provider allocations. If you do remove all allocations for a given server, you can heal them later. See Using heal_allocations for details.

$ openstack --os-placement-api-version 1.12 resource provider allocation set $vm1 \
    --project-id 2f3bffc5db2b47deb40808a4ed2d7c7a \
    --user-id 2206168427c54d92ae2b2572bb0da9af \
    --allocation rp=52d0182d-d466-4210-8f0d-29466bb54feb,VCPU=1 \
    --allocation rp=52d0182d-d466-4210-8f0d-29466bb54feb,MEMORY_MB=512 \
    --allocation rp=52d0182d-d466-4210-8f0d-29466bb54feb,DISK_GB=1
+--------------------------------------+------------+------------------------------------------------+----------------------------------+----------------------------------+
| resource_provider                    | generation | resources                                      | project_id                       | user_id                          |
+--------------------------------------+------------+------------------------------------------------+----------------------------------+----------------------------------+
| 52d0182d-d466-4210-8f0d-29466bb54feb |          4 | {u'VCPU': 1, u'MEMORY_MB': 512, u'DISK_GB': 1} | 2f3bffc5db2b47deb40808a4ed2d7c7a | 2206168427c54d92ae2b2572bb0da9af |
+--------------------------------------+------------+------------------------------------------------+----------------------------------+----------------------------------+

Once the devstack1 resource provider allocations have been removed using either of the approaches above, the devstack1 resource provider can be deleted:

$ openstack resource provider delete $devstack1

And the related compute service if desired:

$ openstack --os-compute-api-version 2.53 compute service delete e3c18c2d-9488-4863-b728-f3f292ec5da8

For more details on the resource provider commands used in this guide, refer to the osc-placement plugin documentation.

Using heal_allocations

If you have a particularly troubling allocation consumer and just want to delete its allocations from all providers, you can use the openstack resource provider allocation delete command and then heal the allocations for the consumer using the heal_allocations command. For example:

$ openstack resource provider allocation delete $vm1
$ nova-manage placement heal_allocations --verbose --instance $vm1
Looking for instances in cell: 04879596-d893-401c-b2a6-3d3aa096089d(cell1)
Found 1 candidate instances.
Successfully created allocations for instance a1e6e0b2-9028-4166-b79b-c177ff70fbb7.
Processed 1 instances.
$ openstack resource provider allocation show $vm1
+--------------------------------------+------------+------------------------------------------------+
| resource_provider                    | generation | resources                                      |
+--------------------------------------+------------+------------------------------------------------+
| 52d0182d-d466-4210-8f0d-29466bb54feb |          5 | {u'VCPU': 1, u'MEMORY_MB': 512, u'DISK_GB': 1} |
+--------------------------------------+------------+------------------------------------------------+

Note that deleting allocations and then relying on heal_allocations may not always be the best solution since healing allocations does not account for some things:

  • Migration-based allocations would be lost if manually deleted during a resize. These are allocations tracked by the migration resource record on the source compute service during a migration.

  • Healing allocations only partially support nested allocations. Nested allocations due to Neutron ports having QoS policies are supported since 20.0.0 (Train) release. But nested allocations due to vGPU or Cyborg device profile requests in the flavor are not supported. Also if you are using provider.yaml files on compute hosts to define additional resources, if those resources are defined on child resource providers then instances using such resources are not supported.

If you do use the heal_allocations command to cleanup allocations for a specific trouble instance, it is recommended to take note of what the allocations were before you remove them in case you need to reset them manually later. Use the openstack resource provider allocation show command to get allocations for a consumer before deleting them, e.g.:

$ openstack --os-placement-api-version 1.12 resource provider allocation show $vm1