Skip to content

Files

Latest commit

 

History

History
353 lines (258 loc) · 15 KB

baremetal-agent-controller.md

File metadata and controls

353 lines (258 loc) · 15 KB

Baremetal Agent Controller (a.k.a BMAC)

BMAC is a Kubernetes controller responsible for reconciling BareMetalHost and Agent (defined and maintained in this repo) resources for the agent-based deployment scenario.

Testing

The testing environment for BMAC consists of

Each of the components listed above provide their own documentation on how to deploy and configure them. However, you can find below a set of recommended configs that can be used for each of these components:

Dev Scripts

# Giving Master nodes some extra CPU since we won't be
# deploying any workers
export MASTER_VCPU=4
export MASTER_MEMORY=20000

# Set specs for workers
export WORKER_VCPU=4
export WORKER_MEMORY=20000
export WORKER_DISK=60

# No workers are needed to test BMAC
export NUM_WORKERS=0

# Add extra workers so we can use it for the deployment.
# SNO requires 1 extra machine to be created.
export NUM_EXTRA_WORKERS=1

# At the time of this writing, this requires the 1195 PR
# mentioned below.
export PROVISIONING_NETWORK_PROFILE=Disabled

# Add extradisks to VMs
export VM_EXTRADISKS=true
export VM_EXTRADISKS_LIST="vda vdb"
export VM_EXTRADISKS_SIZE="30G"

export REDFISH_EMULATOR_IGNORE_BOOT_DEVICE=True

The config above should provide you with an environment that is ready to be used for the operator, assisted installer, and BMAC tests. Here are a few tips that would help simplifying the environment and the steps required:

  • Clone baremetal-operator somewhere and set the BAREMETAL_OPERATOR_LOCAL_IMAGE in your config.

NOTE

The default hardware requirements for the OCP cluster are higher than the values provided below. A guide on how to customize validator requirements can be found here.

Local Baremetal Operator (optional)

NOTE

This section is completely optional. If you don't need to run your own clone of the baremetal-operator, just ignore it and proceed to the next step.

The baremetal-operator will define the BareMetalHost custom resource required by the agent based install process. Setting the BAREMETAL_OPERATOR_LOCAL_IMAGE should build and run the BMO already. However, it's recommended to run the local-bmo script to facilitate the deployment and monitoring of the BMO. Here's what using local-bmo looks like:

It's possible to disable inspection for the master (and workers) nodes before running the local-bmo script. This will make the script less noisy which will make debugging easier.

./metal3-dev/pause-control-plane.sh

The pause-control-plane.sh script only pauses the control plane. You can do the same for the worker nodes with the following command

for host in $(oc get baremetalhost -n openshift-machine-api -o name | grep -e '-worker-'); do
    oc annotate --overwrite -n openshift-machine-api "$host" \
       'baremetalhost.metal3.io/paused=""'
done

The steps mentioned above are optional, and only recommended for debugging purposes. Let's now run local-bmo and move on. This script will tail the logs so do it in a separate buffer so that it can be kept running.

# Note variable is different from the one in your dev-script
# config file. You can set it to the same path, though.
export BAREMETAL_OPERATOR_PATH=/path/to/your/local/clone
./metal3-dev/local-bmo.sh

Assisted Installer Operator

Once the dev-script environment is up-and-running, and the bmo has been deployed, you can proceed to deploying the Assisted Installer Operator. There's a script in the dev-scripts repo that facilitates this step:

[dev@edge-10 dev-scripts]$ ./assisted_deployment.sh install_assisted_service

Take a look at the script itself to know what variables can be customized for the Assisted Installer Operator deployment.

Creating AgentClusterInstall, ClusterDeployment and InfraEnv resources

A number of resources has to be created in order to have the deployment fully ready for deploying OCP clusters. A typical workflow is as follows

NOTE

When deploying AgentClusterInstall for SNO it is important to make sure that machineNetwork subnet matches the subnet used by libvirt VMs (configured by passing EXTERNAL_SUBNET_V4 to the dev-scripts config). It defaults to 192.168.111.0/24 therefore the sample manifest linked above needs to be adapted.

At this moment it's good to check logs and verify that there are no conflicting parameters, the ISO has been created correctly and that the installation can be started once a suitable node is provided.

To check if the ISO has been created correctly, do

oc get infraenv myinfraenv -o jsonpath='{.status.isoDownloadURL}' -n assisted-installer

Creating BareMetalHost resources

The baremetal operator creates the BareMetalHost resources for the existing nodes automatically. For scenarios using extra worker nodes (like SNO), it will be necessary to create BareMetalHost resources manually. Luckily enough, assisted_deployment.sh is one step ahead and it has prepared the manifest for us already.

less ocp/ostest/saved-assets/assisted-installer-manifests/06-extra-host-manifests.yaml

The created BareMetalHost manifest contains already a correct namespace as well as annotations to disable the inspection and cleaning. Below is an example on what it could look like.

Please remember to change the value of the infraenvs.agent-install.openshift.io label in case you are using different than the default one (myinfraenv).

apiVersion: metal3.io/v1alpha1
kind: BareMetalHost
metadata:
  name: ostest-worker-0
  namespace: assisted-installer
  annotations:
    inspect.metal3.io: disabled
  labels:
    infraenvs.agent-install.openshift.io: "myinfraenv"
spec:
  online: true
  bootMACAddress: 00:ec:ee:f8:5a:ba
  automatedCleaningMode: disabled
  bmc:
    address: ....
    credentialsName: bmc-secret

Setting inspect.metal3.io is optional since BMAC will add it automatically. Without it, the BareMetalHost will boot IPA and spend additional time in the inspecting phase when the manifest is applied.

Setting the infraenvs.agent-install.openshift.io is required and it must be set to the name of the InfraEnv to use. Without it, BMAC won't be able to set the ISO Url in the BareMetalHost resource.

It is possible to specify RootDeviceHints for the BareMetalHost resource. Root device hints are used to tell the installer what disk to use as the installation disk. Refer to the baremetal-operator documentation to know more.

NOTE

Prior to ACM 2.9: BMAC sets the field automatedCleaningMode to disabled even if the BareMetalHost manifest specifies another value (e.g. automatedCleaningMode: metadata).

In ACM 2.9+: BMAC no longer sets automatedCleaningMode and will respect the value set in the BareMetalHost manifest when created.

If the automatedCleaningMode field is not set when the BareMetalHost manifest is created, then the BMO's webhook will automatically set the field to metadata.

NOTE

BMAC uses the RootDeviceHints from the BareMetalHost resource to find a matching disk in the corresponding Agent's inventory, and then sets that disk in agent.Spec.InstallationDiskID. If the hints provided to the BMH do not match any disk discovered in the discovery phase, BMAC sets the disk in the Agent's Spec as /dev/not-found-by-hints. This will cause the service to notice that the specified disk does not exist and fail the respective validation, showing the following condition in the Agent resource.

message: 'The Spec could not be synced due to an input error: Requested installation
  disk is not part of the host''s valid disks'
reason: InputError

The log of the service will show the following message indicating that the specified disk (as set by BMAC) is not available. Please note this message is not exposed to the user, but only written in the service log:

level=error msg="failed to set installation disk path </dev/not-found-by-hints> host <29d87175-[...]-eb1efd15fdc0> infra env <d46b1dcd-[...]-2412cd69a0b4>" func="github.com/openshift/assisted-service/internal/bminventory.(*bareMetalInventory).updateHostDisksSelectionConfig" file="/go/src/github.com/openshift/origin/internal/bminventory/inventory.go:5229" error="Requested installation disk is not part of the host's valid disks"

Installation flow

After all the resources described above are created the installation starts automatically. A detailed flow is out of scope of this document and can be found here.

An Agent resource will be created that can be monitored during the installation proces as in the example below

$ oc get agent -A
$ oc get agentclusterinstalls test-agent-cluster-install -o json | jq '.status.conditions[] |select(.type | contains("Completed"))'

After the installation succeeds there are two new secrets created in the assisted-installer namespace

assisted-installer  single-node-admin-kubeconfig  Opaque  1  12h
assisted-installer  single-node-admin-password    Opaque  2  12h

Kubeconfig can be exported to the file with

$ oc get secret single-node-admin-kubeconfig -o json -n assisted-installer | jq '.data' | cut -d '"' -f 4 | tr -d '{}' | base64 --decode > /tmp/kubeconfig-sno.yml

NOTE

ClusterDeployment resource defines baseDomain for the installed OCP cluster. This one will be used in the generated kubeconfig file so it may happen (depending on the domain chosen) that there is no connectivity caused by name not being resolved. In such a scenario a manual intervention may be needed (e.g. manual entry in /etc/hosts).

Troubleshooting

  • I have created the BMH, the ClusterDeployment, and the InfraEnv resources. Why doesn't the node start?

The first thing to do is to verify that an ISO has been created and that it is associated with the BMH. Here are a few commands that can be run to achieve this:

$ oc describe infraenv $YOUR_INFRAENV | grep ISO
$ oc describe bmh $YOUR_BMH | grep Image

NOTE

In case the HUB cluster version is 4.11 or above instead of checking the Image on the BMH you should check the PreprovisioningImage status:

$ oc get preprovisioningimage -n {BMH_NAMESPACE} {BMH_NAME} -ojsonpath={'.status}' | jq
  • InfraEnv's ISO Url doesn't have an URL set

This means something may have gone wrong during the ISO generation. Check the assisted-service logs (and docs) to know what happened.

  • InfraEnv has an URL associated but the BMH Image URL field is not set:

Check that the infraenvs.agent-install.openshift.io label is set in your BareMetalHost resource and that the value matches the name of the InfraEnv's. Remember that both resources must be in the same namespace.

Check that resources in the openshift-machine-api are up and running. cluster-baremetal-operator is responsible for handling the state of the BMH so if that one is not running, your BMH will never move forward.

Check that cluster-baremetal-operator is not configured to ignore any namespaces or CRDs. You can do it by checking the overrides section in

$ oc describe clusterversion version --namespace openshift-cluster-version
  • URL is set everywhere, node still doesn't start

Double check that the BareMetalHost definition has online set to true. BMAC should take care of this during the reconcile but, you know, software, computer gnomes, and dark magic.

  • Node boots but it loooks like it is booting something else

Check that the inspect.metal3.io and automatedCleaningMode are both set to disabled. This will prevent Ironic from doing inspection and any cleaning, which will speed up the deployment process and prevent it from running IPA before running the ISO.

This should be set automatically by BMAC in the part linked here but if that is not the case, start from checking the assisted-service logs as there may be more errors related to the BMH.

NOTE

In case the HUB cluster version is 4.11 or above these annotations shouldn't be set and the IPA is expected to run on the node and register with Ironic. Try checking the BMH custom deploy method is set to start_assisted_install

$ oc get bmh -n  {BMH_NAMESPACE} {BMH_NAME} -ojsonpath={'.spec.customDeploy}' | jq
  • Node boots, but nothing else seems to be happening

Check that an agent has been registered for this cluster and BMH. You can verify this by chekcing the existing agents and find the one that has an interface with a MacAddress that matches the BMH BootMACAddress.

Remember that in between the node booting from the Discovery ISO and the Agent CR being created you may need to wait a few minutes.

NOTE

In case the HUB cluster version is 4.11 or above, IPA is expected to start the assisted agent on the node in order for the node to register. Try checking the BMH custom deploy method is set to start_assisted_install

$ oc get bmh -n  {BMH_NAMESPACE} {BMH_NAME} -ojsonpath={'.spec.customDeploy}' | jq

If there is an agent, the next thing to check is that all validations have passed. This can be done by inspecting the ClusterDeployment and verify that the validation phase has succeeded.