Skip to content

openstack-placement-no-allocation-candidates.md

Latest commit

 

History

History
143 lines (108 loc) · 5.46 KB

openstack-placement-no-allocation-candidates.md

File metadata and controls

143 lines (108 loc) · 5.46 KB
Raw
title OpenStack Placement No Allocation Candidates
slug openstack-placement-no-allocation-candidates
technologies
openstack
placement
severity high
tags
openstack
placement
nova
inventory
allocations
production
related
openstack-nova-no-valid-host-was-found
openstack-cinder-no-weighed-backends-available
last_reviewed 2026-06-27

OpenStack Placement No Allocation Candidates

Error Message

nova-scheduler[2210]: DEBUG nova.scheduler.client.report [req-6d2c...] \
Got no allocation candidates from the placement API. This could be due to insufficient \
resources or a temporary occurrence as compute nodes start up.

$ openstack allocation candidate list --resource VCPU=2 --resource MEMORY_MB=2048
No allocation candidates found.

Description

Placement tracks the inventory of every resource provider (compute nodes, and their nested PCI/PGPU providers) and the allocations consumed against it. The Nova scheduler asks Placement for allocation candidates — providers that can satisfy the requested resource classes and traits — before it ever runs filters. When Placement returns an empty set, the scheduler has nothing to filter and the build fails with NoValidHost. This error means the shortfall is at the inventory/allocation layer, not in Nova's filters.

Technologies

  • openstack (placement, nova-scheduler, nova-compute resource tracker)

Severity

high — no host can be selected for instances needing the requested resources; new builds of the affected shape fail until inventory/allocations are corrected.

Common Causes

  1. Genuine resource exhaustion — no provider has free VCPU, MEMORY_MB, or DISK_GB after allocation ratios and reserved amounts.
  2. Required traits or resource classes (e.g. CUSTOM_*, VGPU, PCI) exist on no provider, or the flavor requests a trait nothing advertises.
  3. Stale allocations from deleted instances pinning capacity ("leaked" allocations).
  4. The compute's resource tracker has not (re)reported inventory after restart or upgrade, so providers show zero/missing inventory.
  5. reserved set equal to total, leaving no allocatable capacity.

Root Cause Analysis

Each nova-compute resource tracker writes its node's inventory (totals, reserved, allocation_ratio) and traits into Placement. On a build, the scheduler issues a GET /allocation_candidates query for the flavor's resources and required traits. Placement returns only providers whose total × allocation_ratio − reserved − used ≥ requested for every class and that carry all required traits. An empty response means either no provider has the headroom, or none advertise a required trait/class — frequently because allocations leaked or inventory was never reported. Comparing inventory against usage per provider isolates which it is.

Diagnostic Commands

# Can Placement satisfy the shape at all?
openstack allocation candidate list --resource VCPU=2 --resource MEMORY_MB=2048

# Providers and their inventory/usage
openstack resource provider list
openstack resource provider inventory list <rp-uuid>
openstack resource provider usage show <rp-uuid>

# Traits a provider advertises (for trait-based requests)
openstack resource provider trait list <rp-uuid>

# Find leaked allocations vs real instances
nova-manage placement audit --verbose

Expected Results

# inventory list
| resource_class | total | reserved | allocation_ratio | used  |
| VCPU           | 64    | 0        | 16.0             | 1024  |  <- fully consumed
| MEMORY_MB      | 128k  | 512      | 1.5              | 191k  |

# audit
Allocation for consumer <uuid> has no matching instance (leaked)

used at the effective cap, a missing required trait, or leaked allocations from the audit each explain the empty candidate set.

Resolution

  1. Real exhaustion: add capacity, free resources, or adjust allocation_ratio / reserved in nova.conf so honest headroom exists.
  2. Leaked allocations: heal them so capacity frees up: 
    nova-manage placement heal_allocations --verbose
  3. Missing inventory: restart nova-compute so the resource tracker re-reports; confirm the provider then lists inventory.
  4. Trait/class mismatch: ensure providers advertise the trait the flavor requires, or fix the flavor's resources:/trait: extra-specs.

Validation

openstack allocation candidate list --resource VCPU=2 --resource MEMORY_MB=2048
# Expect: one or more candidate rows returned
openstack server create --flavor m1.small --image cirros --network private placement-test
openstack server show placement-test -c status -f value   # Expect: ACTIVE

Prevention

  • Monitor per-provider headroom in Placement, not just hypervisor stats.
  • Run nova-manage placement audit periodically to catch leaked allocations.
  • Validate inventory reporting after compute restarts and upgrades.
  • Keep flavor traits/resource-class requests aligned with what providers advertise.

Related Errors

References

Tags

openstack · placement · nova · inventory · allocations · production