Reading the Placement logs, and the Placement lines in Nova's logs¶
Placement holds the inventory of every compute and the allocation of every instance. Its log answers two questions:
did Nova's request reach Placement, and why did Placement refuse it. Nova's side of each exchange is in
nova-scheduler.log (allocation candidates, claims) and nova-compute.log (inventory updates, allocation moves),
and most of Nova's error lines carry Placement's request id, which links the two. Messages are quoted from Placement
and Nova 2026.1.
Fixing allocations that are out of sync is a runbook; this page reads the lines.
Where the log is¶
Kolla-Ansible: /var/log/kolla/placement/placement-api.log (log_file = /var/log/kolla/placement/<service>.log
in placement.conf.j2), and the uWSGI request log placement-api-uwsgi.log, on each controller.
Any: /var/log/placement/ or the journal, depending on the packaging.
The request line¶
Placement logs one line per request at INFO (requestlog.py):
<client address> "<METHOD> <URI>" status: <code> len: <bytes> microversion: <version>
With the oslo.log prefix, it carries Placement's own request id and, for calls from Nova, Nova's request id as the
global request id (Nova's report client passes global_request_id=context.global_id). So the request line of an
allocation write can be found from the instance's create request:
# Kolla-Ansible, on the controllers
sudo grep "<nova-request-id>" /var/log/kolla/placement/placement-api.log
sudo grep -E 'status: (409|400|500)' /var/log/kolla/placement/placement-api.log | tail -20
The useful URIs: GET /allocation_candidates?... (the scheduler asks where the request fits), PUT /allocations/<consumer>
(a claim; the consumer is the instance UUID, or a migration UUID during a move),
PUT /resource_providers/<uuid>/inventories (a compute reports its capacity), DELETE /allocations/<consumer>.
Status codes, and what Placement logs with them¶
| Status on the request line | Placement's own line (message text) | Level | What it means |
|---|---|---|---|
409 on PUT /allocations/ |
Over capacity for <rc> on resource provider <rp>. Needed: <n>, Used: <n>, Capacity: <n> |
WARNING | The claim does not fit: Capacity is (total - reserved) * allocation_ratio. Another instance took the room between scheduling and claim, or the inventory changed. |
409 on PUT /allocations/ |
Allocation for <rc> on resource provider <rp> violates min_unit, max_unit, or step_size. Requested: <n>, min_unit: <n>, max_unit: <n>, step_size: <n> |
WARNING | The flavor asks for more of one resource than a single provider allows in one allocation (max_unit). |
| 409 on allocations or inventories | Exceeded retry limit of <n> on allocations write |
WARNING | Concurrent writes to the same providers kept conflicting (allocation_conflict_retry_count, default 10). |
200 on PUT .../ |
Resource provider <uuid> is now over-capacity for <rc> |
WARNING | A compute reported less capacity than is already allocated (allocation ratio lowered, reserved raised, disk shrank). Existing instances stay; new claims on that resource fail. |
409 on allocations, body detail consumer generation conflict |
— | — | Two writers updated the same consumer's allocations; the second must re-read. Nova retries some of these; when it gives up it raises AllocationUpdateFailed or AllocationMoveFailed, which is what its log shows. |
| 500 | Placement API unexpected error: <exception> with a traceback |
ERROR | An unhandled exception, often the database underneath. |
Sources: placement/objects/allocation.py, placement/objects/resource_provider.py, placement/fault_wrap.py,
placement/conf/placement.py.
The client (Nova) receives the matching error text in the response body, for example Unable to create allocation
for '<rc>' on resource provider '<rp>'. The requested amount would exceed the capacity. or … would violate
inventory constraints. (placement/exception.py).
Placement in Nova's logs¶
nova-scheduler¶
| Line (message text) | Level | What it means |
|---|---|---|
Got no allocation candidates from the Placement API. This could be due to insufficient resources or a temporary occurrence as compute nodes start up. |
INFO | GET /allocation_candidates returned nothing: no provider has the resources and traits requested. Check the request line in Placement's log for the query. |
Source: nova/scheduler/manager.py. What to do next: No valid host with free capacity.
nova-compute (the report client)¶
The compute's resource tracker talks to Placement through Nova's report client (nova/scheduler/client/report.py).
Most of its error lines start with Placement's request id in brackets, which you can grep in placement-api.log:
| Line (message text) | Level | What it means |
|---|---|---|
[<placement-req-id>] Failed to update inventory to [<inventory>] for resource provider with UUID <uuid>. Got <status>: <error text> |
ERROR | Placement refused the compute's inventory. With 409 and allocations in the way, Nova raises InventoryInUse. |
[<placement-req-id>] Failed to retrieve resource provider tree from placement API for UUID <uuid>. Got <status>: <error text>. and the other [<id>] Failed to retrieve … lines |
ERROR | A read from Placement failed. |
[<placement-req-id>] Created resource provider record via placement API for resource provider with UUID <uuid> and name <name>. |
INFO | The compute registered itself (first start, or after its provider was deleted). |
Failed to save allocation for <consumer>. Got HTTP <status>: <text> |
WARNING | An allocation write from the compute side (for example during a move) was refused. |
Unable to post allocations for consumer <uuid> (<status> <text>) / Unable to delete allocation for <type> <uuid>: (<status> <text>) |
WARNING | An allocation could not be written or removed; leftovers become phantom usage. |
Placement API service is not responding. |
WARNING (ERROR at client start) | Nova cannot connect to Placement: the inventory is not refreshed while it lasts. |
Placement service credentials do not work. |
WARNING (ERROR at client start) | The [placement] credentials in nova.conf are refused by Keystone. |
The placement API endpoint was not found. |
ERROR when the client is created, otherwise a rate-limited WARNING | No Placement endpoint in the catalog for the configured interface and region. |
Sources: nova/scheduler/client/report.py.
# On a compute: Placement problems seen by its report client
sudo grep -nE "Placement API service is not responding|Placement service credentials|Failed to update inventory|Failed to save allocation" \
/var/log/kolla/nova/nova-compute.log | tail
# Then follow one placement request id to Placement
sudo grep "<placement-req-id>" /var/log/kolla/placement/placement-api.log
Reading a Placement problem in order¶
- Nova's line first: the scheduler (
Got no allocation candidates) or the compute (Failed to …with a request id in brackets). - Placement's request line for that id: method, URI, status.
- Placement's WARNING next to it: over capacity, unit constraints, retry limit.
- The numbers:
openstack resource provider inventory list <uuid>andopenstack resource provider usage show <uuid>(withosc-placement) against what the line says.
Captured on a lab¶
Kolla-Ansible 22.2.0 (2026.1) all-in-one, 2026-10-27, ids replaced. The create of an instance, found in
placement-api.log by Nova's request id (the first id in brackets is Nova's, the global request id):
INFO placement.requestlog [req-<nova-request-id> req-<placement-request-id> <user-id> <project-id> - - default default] 192.0.2.10 "GET /allocation_candidates?limit=1000&resources=DISK_GB%3A1%2CMEMORY_MB%3A512%2CVCPU%3A1&root_required=..." ...
INFO placement.requestlog [req-<nova-request-id> req-<placement-request-id> ...] 192.0.2.10 "PUT /allocations/<server-id>" status: 204 len: 0 microversion: 1.36
Allocations written by hand for a test consumer with openstack resource provider allocation set (osc-placement),
and an inventory change with openstack resource provider inventory set --amend:
WARNING placement.objects.allocation [...] Over capacity for MEMORY_MB on resource provider <rp-uuid>. Needed: 15400, Used: 512, Capacity: 15480.0
INFO placement.requestlog [...] 192.0.2.10 "PUT /allocations/<consumer>" status: 409 ...
WARNING placement.objects.allocation [...] Allocation for MEMORY_MB on resource provider <rp-uuid> violates min_unit, max_unit, or step_size. Requested: 20000, min_unit: 1, max_unit: 15992, step_size: 1
INFO placement.requestlog [...] 192.0.2.10 "PUT /allocations/<consumer>" status: 409 len: 415 microversion: 1.28
WARNING placement.objects.resource_provider [...] Resource provider <rp-uuid> is now over-capacity for MEMORY_MB
The client got The requested amount would exceed the capacity. (HTTP 409) and ... would violate inventory
constraints. (HTTP 409). After the inventory change (reserved raised by hand in Placement), nova-compute did
not put its own value back within 3 minutes: it was restored by hand.
Validity and limits¶
- Every message is quoted from Placement and Nova 2026.1; Captured on a lab has the lines observed. Not observed:
Exceeded retry limit, consumer generation conflicts, the 500 line, and the report client's lines innova-compute.log. - The 409 for over capacity and for unit constraints was observed; the code for
Exceeded retry limitfollows from the source (ResourceProviderConcurrentUpdateDetected) and was not observed. - The
osc-placementcommands require that client plugin.
Related¶
- Placement allocations out of sync: audit and heal (the fix)
- No valid host with free capacity
- Stale records, phantom allocations
- Where the OpenStack logs are, and how to read one log line
- Logs index
Provenance: written by agent:ozarland-author, from Placement and Nova 2026.1 source code. 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.