No valid host was found in Nova Scheduler after DevStack cycle

DevStack is documented as a development environment rather than a reproducible deployment, and consecutive stack.sh runs can diverge when residual state is left behind. One documented failure condition that can surface in this situation is the Nova Scheduler message No valid host was found, raised when no compute host matches placement, resource, and filter constraints.

The unresolved part is where the residual state lives and how far cleanup must reach. Leftover instance records, stale Placement resource providers, or Neutron network allocations from a prior cycle can each block host selection, and cleanup semantics vary across OpenStack releases and plugin configurations.

Which Placement or Nova records must be cleared for host selection to succeed on an incremental re-run? Is there a documented way to reconcile Placement resource provider state with actual compute inventory after an incomplete unstack? How do operators weigh fast incremental re-runs against a full clean.sh teardown when reproducibility matters?