OpenStack Nova Compute Service

Nova is the compute scheduler and instance lifecycle manager for OpenStack. It receives instance creation requests, selects an appropriate hypervisor through the scheduler and placement service, coordinates with Glance for images, Neutron for networking, and Cinder for volumes, then manages the full instance lifecycle through libvirt/KVM.

Nova operates in a cell-based architecture. Cell0 holds instances that failed to schedule. Cell1 (and beyond) holds running instances. The placement service tracks resource inventories (vCPUs, RAM, disk) independently from Nova, providing accurate capacity data for scheduling decisions.

Deploy

Kolla-Ansible Configuration

globals.yml settings:

# Hypervisor type
nova_compute_virt_type: "kvm"        # Production: hardware KVM
# nova_compute_virt_type: "qemu"     # Nested virt or no VT-x/AMD-V

# Libvirt container
enable_nova_libvirt_container: "yes"

# CPU and RAM allocation ratios
nova_cpu_allocation_ratio: 4.0       # 4:1 vCPU to pCPU (adjust for workload)
nova_ram_allocation_ratio: 1.5       # 1.5:1 vRAM to pRAM
nova_disk_allocation_ratio: 1.0      # 1:1 (do not overcommit disk)

# Console access
nova_console: "novnc"               # Options: novnc, spice
# nova_console: "spice"             # Better performance, less browser support

# Metadata service
enable_nova_metadata: "yes"

# Nova database password
nova_database_password: "{{ vault_nova_database_password }}"
nova_api_database_password: "{{ vault_nova_api_database_password }}"
nova_keystone_password: "{{ vault_nova_keystone_password }}"

Deployment:

# Deploy Nova (as part of full deploy or targeted)
kolla-ansible -i inventory deploy --tags nova

# Verify containers
docker ps --filter "name=nova" --format "table {{.Names}}\t{{.Status}}"
# Expected: nova_api, nova_conductor, nova_scheduler, nova_novncproxy,
#           nova_compute, nova_libvirt (all Up)

Post-deploy verification:

source /etc/kolla/admin-openrc.sh

# Verify compute service registration
openstack compute service list
# All services should show Status=enabled, State=up

# Verify cell mapping
nova-manage cell_v2 list_cells
# Expected: cell0 and cell1 with transport_url and database

# Verify hypervisor
openstack hypervisor list
# Should show each compute node with vCPUs, Memory, Disk

# Verify placement
openstack resource provider list
# Should match hypervisor list

Configure

Flavor Management

Flavors define instance resource profiles. Create a standard set:

# Standard flavors (name, RAM-MB, disk-GB, vCPUs)
openstack flavor create --ram 512   --disk 1   --vcpus 1 m1.tiny
openstack flavor create --ram 2048  --disk 20  --vcpus 1 m1.small
openstack flavor create --ram 4096  --disk 40  --vcpus 2 m1.medium
openstack flavor create --ram 8192  --disk 80  --vcpus 4 m1.large
openstack flavor create --ram 16384 --disk 160 --vcpus 8 m1.xlarge

# Private flavor (restricted to specific project)
openstack flavor create --ram 32768 --disk 320 --vcpus 16 \
  --private m1.xxlarge
openstack flavor set --project dev-team m1.xxlarge

CPU Pinning and NUMA

# nova.conf [DEFAULT] on compute nodes
vcpu_pin_set = 4-15          # Pin guest vCPUs to physical cores 4-15
                              # Reserve cores 0-3 for host OS

# NUMA topology filter
[filter_scheduler]
enabled_filters = ...,NUMATopologyFilter

Flavor extra specs for NUMA:

openstack flavor set m1.large \
  --property hw:cpu_policy=dedicated \
  --property hw:numa_nodes=1

Huge Pages

# On compute host: allocate huge pages
echo 1024 > /sys/kernel/mm/hugepages/hugepages-2048kB/nr_hugepages

# Flavor with huge pages
openstack flavor set m1.large \
  --property hw:mem_page_size=large

Live Migration Configuration

# globals.yml
nova_compute_virt_type: "kvm"
enable_nova_libvirt_container: "yes"

# For live migration, all compute nodes need:
# 1. Shared storage (NFS, Ceph) or block migration
# 2. Same CPU model (or cpu_mode=host-model)
# 3. Matching libvirt versions

# nova.conf [libvirt]
live_migration_tunnelled = false     # Use native QEMU migration
live_migration_uri = qemu+tcp://%s/system
cpu_mode = host-model                # Consistent CPU features across hosts

Scheduler Filters and Weights

# nova.conf [filter_scheduler]
enabled_filters = AvailabilityZoneFilter,ComputeFilter,
  ComputeCapabilitiesFilter,ImagePropertiesFilter,
  ServerGroupAntiAffinityFilter,ServerGroupAffinityFilter,
  AggregateInstanceExtraSpecsFilter,NUMATopologyFilter

# Weight classes determine host preference after filtering
weight_classes = nova.scheduler.weights.ram.RAMWeigher
ram_weight_multiplier = 1.0          # Prefer hosts with more free RAM

VNC Console Configuration

# globals.yml
nova_console: "novnc"

# For external access, set the proxy base URL
nova_novncproxy_base_url: "http://controller:6080/vnc_auto.html"

Operate

Instance Lifecycle

# Create instance
openstack server create --flavor m1.small --image cirros \
  --network private --security-group default my-instance

# Status checks
openstack server show my-instance -c status -c OS-EXT-STS:vm_state

# Power operations
openstack server stop my-instance
openstack server start my-instance
openstack server reboot my-instance          # Soft reboot
openstack server reboot --hard my-instance   # Hard reboot

# Resize (change flavor)
openstack server resize --flavor m1.medium my-instance
openstack server resize confirm my-instance   # After verification
openstack server resize revert my-instance    # If problems

# Live migration (to specific host or auto-select)
openstack server migrate --live controller2 my-instance
openstack server migrate --live-migration my-instance   # Auto-select

# Cold migration
openstack server migrate my-instance

# Evacuate (from failed host)
openstack server evacuate my-instance --host controller2

Compute Node Maintenance

# Disable compute node (stop scheduling new instances)
openstack compute service set --disable \
  --disable-reason "Maintenance window" controller nova-compute

# Migrate all instances off the node
nova host-evacuate-live controller

# After maintenance
openstack compute service set --enable controller nova-compute

Orphan Instance Cleanup

# Find instances in ERROR state
openstack server list --all-projects --status ERROR

# Force delete stuck instances
openstack server delete --force <instance-id>

# Clean orphan ports (instances deleted but ports remain)
openstack port list --device-owner compute:nova --network private
# Cross-reference with server list; delete unattached ports

Aggregate and Availability Zone Management

# Create host aggregate
openstack aggregate create --zone az1 ssd-aggregate
openstack aggregate set --property ssd=true ssd-aggregate
openstack aggregate add host ssd-aggregate compute-node-1

# Flavor targeting aggregate
openstack flavor set m1.ssd --property aggregate_instance_extra_specs:ssd=true

Image Caching

# nova.conf [image_cache]
manager_interval = 2400        # Check for stale images every 40 min
remove_unused_base_images = true
remove_unused_original_minimum_age_seconds = 86400  # 24 hours

Troubleshoot

1. Instance Stuck in BUILD or ERROR

Symptoms: Instance stays in BUILD state or transitions to ERROR immediately.

Diagnosis:

# Check instance fault message
openstack server show <instance-id> -c fault

# Check scheduler logs
docker logs nova_scheduler 2>&1 | tail -50

# Check compute logs
docker logs nova_compute 2>&1 | tail -50

Common causes:

docker logs nova_libvirt

openstack quota show <project>

2. Live Migration Failures

Symptoms: Migration starts but fails partway through, or never begins.

Diagnosis:

openstack server migration list --server <instance-id>
docker logs nova_compute 2>&1 | grep -i "migration\|migrate"

Common causes:

--block-migration

cpu_mode = host-model

3. "No Valid Host" Errors

Symptoms:

openstack server create

Diagnosis:

# Check placement resource inventory
openstack resource provider inventory list <compute-rp-uuid>

# Check resource usage
openstack resource provider usage show <compute-rp-uuid>

# Verify scheduler filters are not eliminating all hosts
docker logs nova_scheduler 2>&1 | grep -i "filter\|no valid"

Fix: Identify which filter eliminated all candidates. Common: insufficient RAM/vCPU/disk in placement, availability zone mismatch, aggregate filter mismatch, anti-affinity constraint.

4. Console Access Failures

Symptoms: VNC console shows "Failed to connect" or blank screen.

Diagnosis:

# Get console URL
openstack console url show my-instance

# Check novncproxy
docker logs nova_novncproxy 2>&1 | tail -20

# Verify port 6080 is open
ss -tlnp | grep 6080

Fix:

• Firewall blocking port 6080:

  firewall-cmd --add-port=6080/tcp --permanent

• Wrong proxy base URL: Check

  nova_novncproxy_base_url

  in globals.yml
• Instance not running: verify

  openstack server show -c status

5. Compute Service Down

Symptoms:

openstack compute service list

Diagnosis:

# Check Nova compute container
docker ps --filter "name=nova_compute"
docker logs nova_compute 2>&1 | tail -50

# Check libvirt
docker logs nova_libvirt 2>&1 | tail -20

# Check hypervisor resources
openstack hypervisor show <hostname>

Fix:

• Container crashed:

  docker restart nova_compute nova_libvirt

• libvirt socket missing: check

  /var/run/libvirt/libvirt-sock

  exists
• KVM module not loaded:

  lsmod | grep kvm

  -- load with

  modprobe kvm_intel

  or

  kvm_amd

• Disk full on compute:

  df -h /var/lib/nova/instances

6. Cell Mapping Issues

Symptoms: Instances exist in database but not visible via API. New instances not appearing.

Diagnosis:

# List cells
nova-manage cell_v2 list_cells
# Check: cell1 has correct transport_url and database connection

# Discover new hosts
nova-manage cell_v2 discover_hosts --verbose

Fix:

• Missing host mapping:

  nova-manage cell_v2 discover_hosts

• Instance in cell0 (scheduling failed): check cell0 database for error records
• Database connection string wrong: verify

  [database]

  in nova.conf matches cell mapping

Integration Points

NASA SE Cross-References

openstack server create