Skip to content

Add a compute node

After building one all-in-one host with the quick start, add a second Rocky Linux 10.2 host as a compute node.

Node 1: all-in-one (controller)
Keystone / Glance / Placement / Nova API
Neutron (OVN) / Horizon
Nova Compute / VM
management network (e.g. 192.0.2.0/24)
Node 2: compute
Nova Compute / libvirt / VM
OVN Controller / OVN Metadata Agent
Item Requirement
OS Rocky Linux 10.2 (x86_64)
Host name a short host name different from the controller’s. It is the Nova host name and the OVN chassis name; if they disagree, ports cannot bind
Virtualisation /dev/kvm (otherwise qemu emulation)
MTU the same management-interface MTU as the controller; a smaller one drops full-size packets in the tunnel
Clock synchronised (chronyd or similar)
RPM set the same RPM set as the controller; compute-add checks and stops otherwise

The second host must reach these TCP ports on the controller’s management IP:

Port Service
5000 Keystone
9292 Glance
8778 Placement
9696 Neutron API
8775 Nova metadata
5672 RabbitMQ
6642 OVN southbound
11211 memcached

Geneve tunnels between the nodes use UDP 6081 in both directions (not checked in advance).

HAGISTACK configures no firewall. If firewalld is active (on some clouds’ Rocky Linux images it is), open the ports yourself:

Terminal window
# on the controller
sudo firewall-cmd --permanent --add-port={5000,9292,8778,9696,8775,5672,6642,11211}/tcp --add-port=6081/udp
# on the compute node
sudo firewall-cmd --permanent --add-port=6081/udp
# on both
sudo firewall-cmd --reload

Limit the sources to the management network.

Clone the same release tag on the second host:

Terminal window
sudo dnf -y install git
git clone --branch rocky10.2-rpms-2026.1 https://github.com/hagix9/hagistack.git

2. Write the credentials on the controller

Section titled “2. Write the credentials on the controller”

In hagistack/rocky10.2 on the controller (node 1):

Terminal window
(umask 077; sudo ./hagistack compute-secrets > compute-secrets.env)
  • Content: the five credentials a compute node needs (RABBIT_PASS, NOVA_SERVICE_PASS, PLACEMENT_SERVICE_PASS, NEUTRON_SERVICE_PASS, METADATA_PROXY_SECRET) and the controller’s RPM-set identity (RPM_SET_SHA256, not a secret). No database password, no admin password.
  • Permissions: your shell, not hagistack, creates the file; umask 077 makes it readable by you only. compute-secrets refuses to write to a terminal.

3. Install the credentials on the compute node

Section titled “3. Install the credentials on the compute node”
Terminal window
# on the controller: copy, then delete
scp compute-secrets.env NODE2:/tmp/
shred -u compute-secrets.env
Terminal window
# on the compute node (node 2)
sudo install -d -m 0700 /etc/hagistack
sudo install -o root -g root -m 0600 /tmp/compute-secrets.env /etc/hagistack/secrets.env
shred -u /tmp/compute-secrets.env

/etc/hagistack/secrets.env is root:root, mode 0600.

In hagistack/rocky10.2 on the compute node:

Terminal window
sudo ./hagistack compute-add --controller-ip CONTROLLER_MGMT_IP

CONTROLLER_MGMT_IP is the controller’s management IP (for example 192.0.2.10).

How the controller got its RPMs Command on the compute node
the published release (github, default) sudo ./hagistack compute-add --controller-ip CONTROLLER_MGMT_IP
built from source (build) on the controller sudo ./hagistack rpm-bundle --output /root/hagistack-rpms.tar, copy it to the node, then sudo ./hagistack compute-add --controller-ip CONTROLLER_MGMT_IP --rpm-source bundle --rpm-bundle /root/hagistack-rpms.tar
your own repository (--repo-url) sudo ./hagistack compute-add --controller-ip CONTROLLER_MGMT_IP --repo-url URL

What compute-add does:

  • It stops if /etc/hagistack/secrets.env is missing or has no valid RPM_SET_SHA256. It never generates secrets.
  • It checks that the controller’s ports are reachable before it changes anything.
  • It stops before installing packages if this node’s RPM set is not the controller’s (the RPM-set check).
  • The node gets no database connection, no admin credential and no provider bridge mapping.

On the controller:

Terminal window
sudo ./hagistack discover-hosts

This maps the new compute host into the Nova cell. The controller also does it every 300 seconds; run the command to use the node at once.

On the controller:

Terminal window
sudo -i
. /etc/hagistack/admin-openrc
openstack compute service list --service nova-compute
openstack hypervisor list

Expected result: two nova-compute lines (node 1 and node 2), both State up and Status enabled; hypervisor list shows both hosts up.

The network side:

Terminal window
openstack network agent list

Node 2’s OVN Controller agent and OVN Metadata agent show Alive = True.

Terminal window
openstack server list
openstack server show demo1 -c name -c status -c OS-EXT-SRV-ATTR:host

OS-EXT-SRV-ATTR:host is the compute host the VM runs on. With the default policy only administrators can see this field.

Per host:

Terminal window
openstack server list --all-projects --host NODE2_HOSTNAME

The Nova scheduler chooses the host:

  1. Filters drop hosts that do not meet the conditions (free RAM, disk, vCPU, service up, …).
  2. Weighers score the remaining hosts; the highest score wins.

So two compute nodes do not mean VMs alternate or spread evenly. HAGISTACK leaves the scheduler at its defaults.

Administrators can request a host at creation time (OpenStack 2026.1 CLI syntax):

Terminal window
# availability zone and host
openstack server create --flavor m1.tiny --image cirros-0.6.3-x86_64 \
--network hagistack-tenant --key-name hagistack-key \
--availability-zone nova:NODE2_HOSTNAME demo2
# compute API 2.74 or later: a requested host
openstack --os-compute-api-version 2.74 server create --flavor m1.tiny \
--image cirros-0.6.3-x86_64 --network hagistack-tenant \
--key-name hagistack-key --host NODE2_HOSTNAME demo3
Symptom What to check
The compute node is not registered run sudo ./hagistack discover-hosts on the controller; sudo ./hagistack status on node 2
nova-compute is down sudo journalctl -u openstack-nova-compute -n 100 on node 2; RabbitMQ (5672) reachable? clocks synchronised?
Not in hypervisor list is it up in openstack compute service list --service nova-compute? was discover-hosts run?
A VM goes to ERROR openstack server show VM -c fault. Refusing to bind port … no OVN chassis for host means the host name and the chassis name disagree: make every host name unique and stable, then re-run
No VM lands on node 2 is node 2’s nova-compute up, with enough free resources? The scheduler does not necessarily spread
The RPM sets differ same source and same tag as the controller? compute-secrets regenerated with the controller’s current version?
compute-add stops with “the controller is not reachable on every port” nothing was installed; check the listed ports on the controller and any firewall in between

See also Troubleshooting.