Skip to content

Latest commit

 

History

History
218 lines (164 loc) · 7.19 KB

standard.md

File metadata and controls

218 lines (164 loc) · 7.19 KB
Raw

Standard

These are my notes on how to deploy a standard OpenStack ControlPlane and DataPlane as described in:

by using my scripts and personal conventions.

VMs

The following VMs are deployed

  • crc: hosts control plane pods
  • edpm-compute-0: standard edpm compute node
  • edpm-compute-1: standard edpm compute node

Prepare

Use deploy.sh sequentially with the following tags set to 1 (unless otherwise indicated other tags are 0).

  • CRC
  • ATTACH
  • PVC and DEPS
  • OPER

Simple Deployment

Set the following to 1 and sequentially run deploy and that's it.

  • CONTROL_SIMPLE
  • EDPM_SIMPLE
  • TEST_SIMPLE

Deployment with Preparation

I don't always use make openstack_deploy or make edpm_deploy because I like to have an unapplied CR file to modify before starting the deployment. In that case I use make *_deploy_prep commands as described here.

Set CONTROL to 1 and while the control plane is still coming up build the EDPM nodes sequentially with the following tags set to 1.

  • EDPM_NODE
  • EDPM_DEPLOY_PREP

When oc get pods shows that nova-api and dnsmasq-dns are running the contorl plane should be ready.

You should then have a working control plane running on crc and edpm-compute-{0,1} will be ready to be configured by Ansible. Use ssh_node.sh for a command to use to SSH into edpm-compute-0.

deploy.sh assumes jumbo frames are enabled on the hypervisor.

Create an OpenStackDataPlaneNodeSet CR

deploy.sh should have created the CR if it was run with EDPM_DEPLOY_PREP enabled.

pushd ~/antelope/crs/
oc create -f data_plane/base/deployment.yaml
popd

The CR comes from running make edpm_deploy_prep from install_yamls. I don't always use make edpm_deploy because I like to have a CR file to review and modify before I start the deployment.

The CR should only contain an OpenStackDataPlane NodeSet and not a Deployment. The CR will be a base for kustomzation overlays.

Create a configuration snippet

Create the libvirt-qemu-nova.yaml configuration snippet (so nested VMs can be booted for testing).

oc create -f snippets/libvirt-qemu-nova.yaml

Create a custom version of the nova service which ships with the operator so that it uses the snippet by adding it to the configMaps list. E.g. here is my version

oc create -f services/dataplane_v1beta1_openstackdataplaneservice_nova_custom.yaml

As per the NOTE in dataplane-operator docs, we cannot redefine a custom version of nova service since the "default service will overwrite the custom service with the same name during role reconciliation".

Create an OpenStackDataPlaneDeployment to run Ansible

The standard ansible deployment contains the list of services to be configured including the nova-custom service which uses the configuration snippet from the previous step.

oc create -f deployments/deployment-standard.yaml

I then like to use the following to watch the playbooks:

oc get pods -w -l app=openstackansibleee
oc logs -f dataplane-deployment-configure-network-deployment-standardsvs8r

Stop Failing Ansible Jobs

If Ansible fails and you want to tell the dataplane-operator to stop spawning Ansible jobs, then delete the OpenStackDataPlaneNodeDeployment CR.

oc delete -f deployments/deployment-standard.yaml

You can then update OpenStack DataPlane NodeSet (e.g. data_plane/base/deployment.yaml) or Deployment (e.g. deployments/deployment-standard.yaml) recreate the deployment to try again.

The above shouldn't be necessary unless the Ansible jobs have a problem. If all of the Ansible jobs succeed, then the completed jobs (returned when running oc get pods | grep edpm) will eventually be terminated.

Check if the configuration snipet was applied

After the compute node is deployed, use the following to confirm if the configuration snippet was copied to the host.

$(./ssh_node.sh)
ls /var/lib/openstack/config/nova/
podman exec -ti nova_compute ls -l /etc/nova/nova.conf.d/

The above should show a configuration file from the configuration snippet (created earlier) in the nova_compute pod's nova.conf.d directory.

Test

Use test.sh with:

  • CEPH and CINDER set to 0 (there's no backend for Cinder)

You should be able to create a glance image (hosted on a local storage provided by a PVC), a private network and be able to boot a VM.

Clean

Delete Deployment

Delete the OpenStackDataPlaneDeployment CR.

oc delete -f deployments/deployment-standard.yaml

Another way to do this if you don't have the CR file is.

oc delete openstackdataplanedeployments.dataplane.openstack.org/deployment-standard

Use openstackdataplanedeployments.dataplane.openstack.org to determine the name.

Delete NodeSet

Delete the OpenStackDataPlaneNodeSet CR.

oc delete -f data_plane/base/deployment.yaml

Another way to do this if you don't have the CR file is.

oc delete openstackdataplanenodesets.dataplane.openstack.org/openstack-edpm

Use oc edit openstackdataplanenodesets.dataplane.openstack.org to determine the name.

clean.sh script

Use clean.sh. The script deletes all found OpenStackDataPlaneDeployments and OpenStackDataPlaneNodeSets.

  • Set EDPM_NODE, CONTROL, and PVC to 1

The script only needs to be run once with all of the above. Other vars can keep their defualts of 0 (though NODES defaults to 2).

Deploy Again

  • To deploy again without having to redploy and configure CRC, use rebuild.sh.

  • When oc get pods shows that nova-api and dnsmasq-dns are running the control plane should be ready.

  • You should now be able to start again from "Create an OpenStackDataPlaneNodeSet CR"

Clean Everything

Use clean.sh.

  • Set OPERATORS and CRC to 1 to remove everything