A compute was renamed, reinstalled or re-added. nova-compute refuses to start, or gets no resource provider¶
Nova ties a compute to three records that must agree: the service (by host name), the compute node (by a UUID that
the compute keeps on disk in a compute_id file), and the Placement resource provider (same UUID, named after the
hypervisor host name). When the host name changes, the compute_id file is lost, or a compute is deleted and added
back under the same name while its old provider stays in Placement, one of them disagrees. Then nova-compute
refuses to start, or starts but never gets a resource provider, so nothing is scheduled on it.
Symptom¶
One of these in the nova-compute log of the compute (Kolla-Ansible: /var/log/kolla/nova/nova-compute.log).
The messages are from nova/compute/manager.py and nova/scheduler/client/report.py.
The service refuses to start. On the Kolla-Ansible 2026.1 lab, for Possible rename detected, nova-compute logged
the exception and kept running, without heartbeat: the container stayed Up (healthy), with no restart loop,
and the service went down. Look in the log, not at the container status:
... nova.exception.InvalidConfiguration: My node <node-uuid> has host '<old-host>' but my host is '<new-host>'; Possible rename detected, refusing to start!
... nova.exception.InvalidConfiguration: My compute node <node-uuid> has hypervisor_hostname <old-hypervisor-hostname> but virt driver reports it should be <new-hypervisor-hostname>. Possible rename detected, refusing to start!
... nova.exception.InvalidConfiguration: No local node identity found, but this is not our first startup on this host. Refusing to start after potentially having lost that state!
... nova.exception.InvalidConfiguration: My hypervisor has existing instances, but I appear to be a new service in this database. Possible database configuration error, refusing to start!
Or the service starts and is up, but every resource update fails because Placement refuses the provider:
ERROR nova.scheduler.client.report [...] [req-<placement-request-id>] Failed to create resource provider record in placement API for UUID <new-node-uuid>. Got 409: {"errors": [{"status": 409, "title": "Conflict", "detail": "... Conflicting resource provider name: <hypervisor-hostname> already exists. ", ...}]}.
ERROR nova.compute.manager [...] Error updating resources for node <hypervisor-hostname>.
...
nova.exception.ResourceProviderCreationFailed: Failed to create resource provider <hypervisor-hostname>
The provider list then shows the name with the old UUID, which no hypervisor has any more:
$ openstack resource provider list --name <hypervisor-hostname> -c uuid -c name
| <old-node-uuid> | <hypervisor-hostname> |
$ openstack hypervisor list --long -c ID -c "Hypervisor Hostname" -c State
| <new-node-uuid> | <hypervisor-hostname> | up |
On Kolla-Ansible, a deploy can also stop before that, in the nova-cell role:
"Found multiple compute services with same host name, please check your Nova compute agents list and fix it. The query with hypervisor name '<nodename>' returned following: [...]"
Impact¶
nova-computerefusing to start: no new instance, migration or resize on that compute, and the compute's service goesdownafterservice_down_time. Its running instances keep running.- No resource provider: the compute is
upbut never in the scheduler's candidates. Builds that need it fail withNo valid host, its capacity is lost; nothing else breaks. - Running instances are not stopped by any of these. They become a problem only if someone "fixes" the compute by deleting records while they still run on it.
Quick checks¶
Read-only. With admin credentials, plus a shell on the compute.
# 1. What Nova knows: the service and the compute node for this host, and their state
openstack compute service list --service nova-compute --host <host> --long
openstack --os-compute-api-version 2.53 hypervisor list --long -c ID -c "Hypervisor Hostname" -c State
# with compute API 2.53 or later, the hypervisor ID is the compute node UUID, which is the provider UUID
# 2. What Placement knows for the same name
openstack resource provider list --name <hypervisor-hostname>
# 3. What the compute itself says it is
hostname; hostname -f
# Kolla-Ansible: the identity file inside the container, and the one the deploy wrote for it
sudo docker exec nova_compute cat /var/lib/nova/compute_id
sudo cat /etc/kolla/nova-compute/compute_id
# Any: /var/lib/nova/compute_id (the [DEFAULT] state_path), or compute_id next to nova.conf
# 4. Does anything run there, and does Placement hold allocations on the old provider?
openstack server list --all-projects --host <host> -c ID -c Status
openstack resource provider show --allocations <old-node-uuid>
Expected on a healthy compute: one service, one hypervisor whose ID equals the content of compute_id, and one
provider with that same UUID and the hypervisor host name.
Diagnosis¶
How Nova identifies a compute (2023.1 and later)¶
From the upstream Compute Node Identification guide and nova/virt/node.py:
- At first start,
nova-computegenerates a UUID and writes it tocompute_idin[DEFAULT] state_path(/var/lib/nova), or uses acompute_idfile found next to one of its configuration files. It reads this file at every start; this UUID is the compute node's identity, and the Placement provider's UUID. - The service record is still matched by
[DEFAULT] host, which defaults to the system host name. Kolla-Ansible does not sethostfor libvirt computes, so the system host name is what counts. - At start-up (
init_hostinnova/compute/manager.py),nova-computerefuses to start when:- the compute node of its
compute_idbelongs to another host name (Possible rename detected); - its service already existed but no
compute_idfile is found (No local node identity found); - its service is new but libvirt already runs instances (
My hypervisor has existing instances).
- the compute node of its
- The upstream guide says the host name must not change after the compute registered, even with the identity file.
How Kolla-Ansible writes compute_id¶
The nova-cell role (tasks/config.yml, Ensure compute_id exists in nova_compute volume) lists the hypervisors,
matches the one whose name is the host's ansible_facts.nodename, and writes its ID to
/etc/kolla/nova-compute/compute_id; the container copies it to /var/lib/nova/compute_id at start
(nova-compute.json.j2). If no hypervisor matches, it writes a UUID derived from the node name. Two consequences:
- a reinstalled compute that keeps its name gets its old UUID back at the next deploy;
- a renamed compute gets a new UUID, so a new compute node and a new provider, while the old ones stay.
Where the provider name conflict comes from¶
Placement requires provider names to be unique. A new compute node UUID with the same hypervisor host name as an
existing provider is refused with 409 Conflicting resource provider name. The old provider is left behind when:
- the compute service was deleted but its provider was not:
openstack compute service deletedeletes the provider and its allocations, but ignores a provider that still has allocations afterwards (ResourceProviderInUseis swallowed indelete_resource_provider). The only trace is the 409 that_delete_providerlogs in thenova-apilog:Failed to delete resource provider with UUID <uuid> from the placement API. Got 409: .... When Placement cannot be reached at all, thenova-apilog hasFailed to delete compute node resource provider for compute node <uuid>: ...instead; - the compute's identity changed (a new
compute_id) while its old compute node and provider stayed.
Possible causes¶
| What you see | Cause |
|---|---|
Possible rename detected (the node's host, or its hypervisor host name, changed) |
Cause 1: the host name changed |
My hypervisor has existing instances, but I appear to be a new service |
Cause 1 too: the new name created a new service, while the instances run under the old one |
No local node identity found |
Cause 2: the compute_id file was lost (reinstall) |
Conflicting resource provider name, one provider with the old UUID and no hypervisor with that ID |
Cause 3: a provider left over from a deleted or re-identified compute |
Kolla-Ansible: Found multiple compute services with same host name |
Two compute services, with different Host values (typically a short name and an FQDN), report the same hypervisor host name; check Quick checks 1 |
Cause 1: the host name changed¶
hostname on the compute differs from the Host of its service in compute service list. Typical after an OS
reinstall with a new name, a change from a short name to an FQDN or back, or a change in DNS or /etc/hosts that
changes what hostname returns.
Kolla-Ansible: the nova_compute container keeps the host name it was created with. On the lab, hostnamectl
set-hostname followed by a restart of the container did not change the name nova-compute saw (docker exec
nova_compute hostname still gave the old one). Whether a deploy that recreates the container brings the new name
in was not tested.
Cause 2: the identity file was lost¶
Same host name, the service exists, but compute_id is missing in the container. On Kolla-Ansible, the container
copies /etc/kolla/nova-compute/compute_id into /var/lib/nova/compute_id at every start, so removing the
nova_compute volume alone does not cause this: the file in /etc/kolla/nova-compute/ must be lost too, for
example after an OS reinstall, and the deploy has not run since. Then the container itself does not start, before
nova-compute runs: kolla_start logs ERROR MissingRequiredSource: /var/lib/kolla/config_files/compute_id file is
not found (docker logs nova_compute) and systemd restarts it in a loop. The No local node identity found
message comes from deployments that do not copy the file in.
Cause 3: an old provider with the same name¶
openstack resource provider list --name <hypervisor-hostname> returns a UUID that hypervisor list does not have.
Resolution¶
For cause 1: give the compute its old name back¶
This is the only change that does not touch Nova's records, and the upstream guide does not support renaming a registered compute.
⚠️ State-changing. Changes the system host name of the compute. Anything else on the host that uses the name (monitoring, certificates, the deployment inventory) sees the old name again. Precondition: the old name is the
Hostof the existing service incompute service list. Undo: set the new name again.
sudo hostnamectl set-hostname <old-host>
Kolla-Ansible: check that the inventory uses that name too, then redeploy the compute so that compute_id and
the configuration are written again.
⚠️ State-changing. Rewrites the compute's Nova configuration and
compute_id, and restarts its Nova containers. Running instances keep running. Precondition: the host name and the inventory are the old ones again. Undo: none needed.
kolla-ansible reconfigure -i <inventory> -t nova --limit <old-host>
Any: restart nova-compute.
If nova-compute started under the new name before you noticed, clean up what it left:
- on Kolla-Ansible, a
compute_idthat a deploy run under the new name rewrote with a UUID derived from it: thereconfigureabove writes the old hypervisor's ID back once the name and the inventory are the old ones again. After the reconfigure, check that/etc/kolla/nova-compute/compute_idandsudo docker exec nova_compute cat /var/lib/nova/compute_idboth equal the old hypervisor ID, and that the service isup; - a second compute service with the new name,
downnow, with its compute node and provider, if any. Delete it after checking that no instance is listed under that name:
🛑 Destructive. Deletes the second service record, its compute node and its provider, if any. What can be lost: the record of the real compute if the wrong ID is taken. Precondition: the service to delete is the one whose
Hostis the new name, it isdown, andopenstack server list --all-projects --host <new-host>is empty. Undo: none.
openstack compute service list --service nova-compute --long
openstack compute service delete <service-id-of-new-host>
If the compute must really have a new name, treat it as a new compute: move its instances off
(maintenance), delete its compute service (see
cause 3 for checking that the provider went with it), rename, deploy it as a new compute, and map it with
nova-manage cell_v2 discover_hosts.
For cause 2: put the old identity back¶
The UUID is the hypervisor ID of this host in Quick checks 1 (compute API 2.53 or later).
Kolla-Ansible: a deploy writes it back:
⚠️ State-changing. Rewrites the compute's configuration and restarts its containers. Running instances keep running (see restarting nova_libvirt). Precondition: one hypervisor in
hypervisor listwith this host's node name. Undo: none needed.
kolla-ansible reconfigure -i <inventory> -t nova --limit <host>
Any: write it by hand, then start nova-compute.
⚠️ State-changing. Gives the compute its identity back. Precondition: the UUID is the one of this host's hypervisor, not a new one; a wrong UUID makes
nova-computetake over another compute's records. Undo: remove the file.
openstack --os-compute-api-version 2.53 hypervisor list -f value -c ID -c "Hypervisor Hostname" | grep -w <hypervisor-hostname>
echo -n <node-uuid> | sudo tee /var/lib/nova/compute_id
sudo chown nova:nova /var/lib/nova/compute_id
For cause 3: remove the old provider¶
First prove it is the leftover: its UUID is not the ID of any hypervisor (Quick checks 1), and no instance in Nova runs on it.
🛑 Destructive. Deletes the old provider, and before that its allocations. What can be lost: the allocations it holds. For an instance that has moved to another compute, only the part on the old provider goes. For an instance that still runs on this compute, its only allocation goes: it has none until it is healed on the new provider (last step below), and until then the scheduler overcommits the compute. Precondition: every migration in a final state and no task state on the instances of this host, and the old provider's allocations saved. Undo: none for the provider; the new compute node creates its own provider by itself.
# save what it holds
openstack resource provider show --allocations <old-node-uuid> -f json > provider-<old-node-uuid>.json
# remove its allocations. The old compute node no longer exists, so audit prints
# "The compute node for UUID <old-node-uuid> can not be found" and lists every consumer on that provider.
sudo docker exec nova_api nova-manage placement audit --verbose --resource_provider <old-node-uuid>
sudo docker exec nova_api nova-manage placement audit --verbose --delete --resource_provider <old-node-uuid>
# then the provider itself
openstack resource provider delete <old-node-uuid>
Any: the same nova-manage placement audit commands where nova.conf has [api_database] and [placement].
See Placement allocations out of sync for what audit
reports and its preconditions.
At its next resource update (every 60 seconds by default), nova-compute creates its provider: Created resource
provider record via placement API for resource provider with UUID <new-node-uuid> and name <hypervisor-hostname>.
Restarting nova-compute does the same at once.
Then give the instances that run on this compute their allocations back, on the new provider:
⚠️ State-changing. Creates allocations from each instance's flavor on the new provider. Precondition: the new provider exists (
resource provider list --name <hypervisor-hostname>shows<new-node-uuid>). Undo:openstack resource provider allocation delete <server-id>.
For each consumer in the saved provider-<old-node-uuid>.json that is an instance on this compute:
sudo docker exec nova_api nova-manage placement heal_allocations --verbose --dry-run --instance <server-id>
sudo docker exec nova_api nova-manage placement heal_allocations --verbose --instance <server-id>
If two compute services report the same hypervisor host name (the Kolla-Ansible assert), they have different
Host values, typically a short name and an FQDN. Delete the one that is down and has no instance, after checking
its ID and running the instance check with its own Host value:
🛑 Destructive. Deletes the compute service record, its compute node, its host mapping, its aggregate memberships and its provider. What can be lost: the record of the compute if the wrong ID is taken. Precondition: the service is
down, andopenstack server list --all-projects --host <that-service-host>is empty (Nova refuses otherwise:Unable to delete compute service that is hosting instances). Undo: none; a compute that starts again under that name registers anew.
openstack compute service delete <service-id>
Verification¶
# one service, up and enabled
openstack compute service list --service nova-compute --host <host>
# one hypervisor, whose ID is the compute's compute_id
openstack --os-compute-api-version 2.53 hypervisor list --long -c ID -c "Hypervisor Hostname" -c State
sudo docker exec nova_compute cat /var/lib/nova/compute_id
# one provider with that UUID and the hypervisor host name, with its inventory
openstack resource provider list --name <hypervisor-hostname>
openstack resource provider inventory list <node-uuid>
# the compute is a candidate again
openstack allocation candidate list --resource VCPU=1 --resource MEMORY_MB=512 | grep <node-uuid>
No more Error updating resources for node or Failed to create resource provider in the nova-compute log at the
next resource update.
Lab observations¶
OpenStack 2026.1, Kolla-Ansible 22.2.0, 3 controllers and 2 computes, Ubuntu 24.04, ML2/OVS, 2026-10-27, on a compute without instances.
- Quick checks 1 to 4 ran as written on a healthy compute: one service, one hypervisor whose ID is the content of
both
compute_idfiles, one provider with that UUID. - Cause 1, simulated with
host = <new-host>in[DEFAULT]of the compute'snova.conf(the system rename alone did not reach the container, see cause 1):My node <node-uuid> has host '<old-host>' but my host is '<new-host>'; Possible rename detected, refusing to start!, logged afterStarting compute node; the process stayed up, the containerhealthy, no new service record. Removing the line and restarting the container gave the service backup. - Cause 2, both
compute_idfiles removed: the container exits withMissingRequiredSourceand restarts in a loop.kolla-ansible reconfigure -i <inventory> -t nova --limit <host>(2 minutes) ran Copy the matched hypervisor UUID to compute_id and the service came backup, both files with the old hypervisor ID. - Cause 3 not reproduced.
openstack compute service deletewith an allocation of an unknown consumer still on the provider left the provider (nova-api:Failed to delete resource provider with UUID <uuid> from the placement API. Got 409), as the page says, and also left the compute node record. With a new UUID incompute_id,nova-computethen refused to start withUnable to create duplicate record for compute node <host>:<new-uuid>andInvalidConfiguration: Duplicate compute node record found for host <host> node <host>, not with the Placement 409. Recovered by deleting the allocation, writing the old UUID back and runningnova-manage cell_v2 discover_hosts(the service delete had removed the host mapping). - Not run: the resolutions of cause 3 (audit, provider delete, heal), the service delete of a second service, the
Kolla-Ansible
Found multiple compute servicesassert, a real OS reinstall or rename, and the hand-writtencompute_idof Any.
Validity and limits¶
- Written from Nova 2026.1 (
nova/compute/manager.py:init_host,_ensure_existing_node_identity,_check_for_host_rename,_sanity_check_new_host,_update_available_resource_for_node;nova/virt/node.py;nova/scheduler/client/report.py:_create_resource_provider,delete_resource_provider;nova/api/openstack/compute/services.py: service delete), Placement 2026.1 (placement/handlers/resource_provider.py, the duplicate name message), the upstream Compute Node Identification guide, and Kolla-Ansible 2026.1 (nova-cellrole,tasks/config.ymlandnova-compute.json.j2). Causes 1 and 2 reproduced on a Kolla-Ansible lab, cause 3 not: see Lab observations. - The log lines are shown with the logger and message from the source; the exception lines are the last line of the traceback that the service prints (on the lab, the process stayed up after printing it). The JSON in the Placement 409 is trimmed.
- Not covered: Ironic (no
compute_id, nodes move between services by design), computes with more than one node, and renaming a compute while its instances stay on it, which upstream does not support.
Related¶
- Placement allocations do not match the instances
- No valid host was found, although capacity exists: an unmapped host, and
discover_hosts. - Moving instances off a compute host for planned maintenance
- Nova 2026.1: Compute Node Identification
Provenance: written by agent:ozarland-author, from the Nova, Placement and Kolla-Ansible 2026.1 documentation and source. Reviewed in its pull request (see the page history). Checked by a human operator: not yet.
Worked for you, or not? Tell us in one line: your release, your deployment method, and what happened.