id: PG-02
category: vm-lifecycle
title: No valid host, with free capacity?
situation: >-
  Builds end in `ERROR` with no host assigned, while computes have free CPU, memory and disk. The request
  stopped before any compute was asked.

diagram:
  type: branches
  head: [The fault says, The request stopped at]
  split: 47
  rows:
    - see: "No valid host was found. There are not enough hosts available."
      mono: true
      means: A scheduler filter
      sub: "Scheduler log: `Filter <Name> returned 0 hosts`"
      tone: warn
    - see: "No valid host was found."
      mono: true
      means: Placement, no candidate
      sub: Inventory, ratios, reserved, or a disabled compute
      tone: warn
    - see: "Host '<host>' is not mapped to any cell"
      mono: true
      means: The cell mapping
      sub: The host exists but is not mapped
      tone: stop
  note: Message variants as observed on a lab on 2026.1. Running instances are not affected.

checks:
  - label: Computes up and enabled?
    hint: openstack compute service list --service nova-compute
  - label: Every host mapped to a cell?
    hint: nova-manage cell_v2 list_hosts
  - label: Placement candidates for the flavor
    hint: openstack allocation candidate list --resource VCPU=<n> ...
  - label: What the scheduler logged
    hint: nova-scheduler.log for the request id of the failed boot

caution:
  - level: state-changing
    text: >-
      Re-enabling a compute makes it schedulable again. Read its disable reason first: it may still apply.
  - level: avoid
    text: >-
      Changing a ratio directly in Placement. The compute usually overrides it on its next update: set ratios and
      reserved values in `nova.conf`.

lesson: Three places a build can stop before any compute. Read the exact message first.

runbook:
  path: scheduling/no-valid-host-with-free-capacity.md
  status: lab-verified
  release: "2026.1"

scope: >-
  Cells v2 and Placement apply to every current release. Lab-verified on OpenStack 2026.1 (all-in-one,
  Kolla-Ansible): unmapped host, reserved memory, disabled service and a filter. Stale allocations and missing
  traits were not reproduced. A plain allocation candidate list still shows a disabled compute: add
  --forbidden COMPUTE_STATUS_DISABLED (Placement API 1.22) to see what Nova sees. The allocation candidate commands need the osc-placement plugin. Ironic is not
  covered.

linkedin:
  title: "\"No valid host\": three places the request can stop"
  caption: |
    Capacity exists. Hypervisors have free CPU, memory and disk. And every build still ends with "No valid host was found".

    Before opening a capacity ticket, read the exact message. It tells you where the request stopped:

    → "...There are not enough hosts available." A scheduler filter rejected every host. The scheduler log names it.
    → "No valid host was found." and nothing after: Placement returned no candidate. Look at inventory, allocation ratios, reserved values, disabled computes.
    → "Host '<host>' is not mapped to any cell": the compute exists but was never mapped. Classic after adding hosts.

    Three places the request can stop. And the third message does not even say "No valid host".
  hashtags: [OpenStack, Nova, Placement, CloudOperations, SRE, OzarLand]
