Troubleshooting

Troubleshooting the installer workflow
Troubleshooting install-config.yaml
Bootstrap VM issues
- Bootstrap VM cannot boot up the cluster nodes
- Inspecting logs
Cluster nodes will not PXE boot
The API is not accessible
Cleaning up previous installations
Issues with creating the registry
Miscellaneous issues
Reviewing the installation

Troubleshooting the installer workflow

Prior to troubleshooting the installation environment, it is critical to understand the overall flow of the installer-provisioned installation on bare metal. The diagrams below provide a troubleshooting flow with a step-by-step breakdown for the environment.

Flow-Diagram-1

Workflow 1 of 4 illustrates a troubleshooting workflow when the install-config.yaml file has errors or the Fedora CoreOS (FCOS) images are inaccessible. Troubleshooting suggestions can be found at Troubleshooting install-config.yaml.

Flow-Diagram-2

Workflow 2 of 4 illustrates a troubleshooting workflow for bootstrap VM issues, bootstrap VMs that cannot boot up the cluster nodes, and inspecting logs. When installing a OKD cluster without the provisioning network, this workflow does not apply.

Flow-Diagram-3

Workflow 3 of 4 illustrates a troubleshooting workflow for cluster nodes that will not PXE boot. If installing using RedFish Virtual Media, each node must meet minimum firmware requirements for the installer to deploy the node. See Firmware requirements for installing with virtual media in the Prerequisites section for additional details.

Flow-Diagram-4

Workflow 4 of 4 illustrates a troubleshooting workflow from a non-accessible API to a validated installation.

Troubleshooting `install-config.yaml`

The install-config.yaml configuration file represents all of the nodes that are part of the OKD cluster. The file contains the necessary options consisting of but not limited to apiVersion, baseDomain, imageContentSources and virtual IP addresses. If errors occur early in the deployment of the OKD cluster, the errors are likely in the install-config.yaml configuration file.

Procedure

Use the guidelines in YAML-tips.
Verify the YAML syntax is correct using syntax-check.
Verify the Fedora CoreOS (FCOS) QEMU images are properly defined and accessible via the URL provided in the install-config.yaml. For example:
```
$ curl -s -o /dev/null -I -w "%{http_code}\n" http://webserver.example.com:8080/rhcos-44.81.202004250133-0-qemu.x86_64.qcow2.gz?sha256=7d884b46ee54fe87bbc3893bf2aa99af3b2d31f2e19ab5529c60636fbd0f1ce7
```
If the output is 200, there is a valid response from the webserver storing the bootstrap VM image.

Bootstrap VM issues

The OKD installation program spawns a bootstrap node virtual machine, which handles provisioning the OKD cluster nodes.

Procedure

About 10 to 15 minutes after triggering the installation program, check to ensure the bootstrap VM is operational using the virsh command:
```
$ sudo virsh list
```
```
 Id    Name                           State
 --------------------------------------------
 12    openshift-xf6fq-bootstrap      running
```
The name of the bootstrap VM is always the cluster name followed by a random set of characters and ending in the word "bootstrap."

If the bootstrap VM is not running after 10-15 minutes, troubleshoot why it is not running. Possible issues include:

Verify libvirtd is running on the system:

$ systemctl status libvirtd

● libvirtd.service - Virtualization daemon
   Loaded: loaded (/usr/lib/systemd/system/libvirtd.service; enabled; vendor preset: enabled)
   Active: active (running) since Tue 2020-03-03 21:21:07 UTC; 3 weeks 5 days ago
     Docs: man:libvirtd(8)
           https://libvirt.org
 Main PID: 9850 (libvirtd)
    Tasks: 20 (limit: 32768)
   Memory: 74.8M
   CGroup: /system.slice/libvirtd.service
           ├─ 9850 /usr/sbin/libvirtd

If the bootstrap VM is operational, log in to it.

Use the virsh console command to find the IP address of the bootstrap VM:

$ sudo virsh console example.com

Connected to domain example.com
Escape character is ^]
Red Hat Enterprise Linux CoreOS 43.81.202001142154.0 (Ootpa) 4.3
SSH host key: SHA256:BRWJktXZgQQRY5zjuAV0IKZ4WM7i4TiUyMVanqu9Pqg (ED25519)
SSH host key: SHA256:7+iKGA7VtG5szmk2jB5gl/5EZ+SNcJ3a2g23o0lnIio (ECDSA)
SSH host key: SHA256:DH5VWhvhvagOTaLsYiVNse9ca+ZSW/30OOMed8rIGOc (RSA)
ens3:  fd35:919d:4042:2:c7ed:9a9f:a9ec:7
ens4: 172.22.0.2 fe80::1d05:e52e:be5d:263f
localhost login:

When deploying a OKD cluster without the provisioning network, you must use a public IP address and not a private IP address like 172.22.0.2.

After you obtain the IP address, log in to the bootstrap VM using the ssh command:

In the console output of the previous step, you can use the IPv6 IP address provided by ens3 or the IPv4 IP provided by ens4.
```
$ ssh core@172.22.0.2
```

If you are not successful logging in to the bootstrap VM, you have likely encountered one of the following scenarios:

You cannot reach the 172.22.0.0/24 network. Verify the network connectivity between the provisioner and the provisioning network bridge. This issue might occur if you are using a provisioning network. `
You cannot reach the bootstrap VM through the public network. When attempting to SSH via baremetal network, verify connectivity on the provisioner host specifically around the baremetal network bridge.
You encountered Permission denied (publickey,password,keyboard-interactive). When attempting to access the bootstrap VM, a Permission denied error might occur. Verify that the SSH key for the user attempting to log into the VM is set within the install-config.yaml file.

Bootstrap VM cannot boot up the cluster nodes

During the deployment, it is possible for the bootstrap VM to fail to boot the cluster nodes, which prevents the VM from provisioning the nodes with the FCOS image. This scenario can arise due to:

A problem with the install-config.yaml file.
Issues with out-of-band network access when using the baremetal network.

To verify the issue, there are three containers related to ironic:

ironic-api
ironic-conductor
ironic-inspector

Procedure

Log in to the bootstrap VM:
```
$ ssh core@172.22.0.2
```
To check the container logs, execute the following:
```
[core@localhost ~]$ sudo podman logs -f <container-name>
```
Replace <container-name> with one of ironic-api, ironic-conductor, or ironic-inspector. If you encounter an issue where the control plane nodes are not booting up via PXE, check the ironic-conductor pod. The ironic-conductor pod contains the most detail about the attempt to boot the cluster nodes, because it attempts to log in to the node over IPMI.

Potential reason

The cluster nodes might be in the ON state when deployment started.

Solution

Power off the OKD cluster nodes before you begin the installation over IPMI:

$ ipmitool -I lanplus -U root -P <password> -H <out-of-band-ip> power off

Inspecting logs

When experiencing issues downloading or accessing the FCOS images, first verify that the URL is correct in the install-config.yaml configuration file.

Example of internal webserver hosting FCOS images

bootstrapOSImage: http://<ip:port>/rhcos-43.81.202001142154.0-qemu.x86_64.qcow2.gz?sha256=9d999f55ff1d44f7ed7c106508e5deecd04dc3c06095d34d36bf1cd127837e0c
clusterOSImage: http://<ip:port>/rhcos-43.81.202001142154.0-openstack.x86_64.qcow2.gz?sha256=a1bda656fa0892f7b936fdc6b6a6086bddaed5dafacedcd7a1e811abb78fe3b0

The ipa-downloader and coreos-downloader containers download resources from a webserver or the external quay.io registry, whichever the install-config.yaml configuration file specifies. Verify the following two containers are up and running and inspect their logs as needed:

ipa-downloader
coreos-downloader

Procedure

Log in to the bootstrap VM:
```
$ ssh core@172.22.0.2
```
Check the status of the ipa-downloader and coreos-downloader containers within the bootstrap VM:
```
[core@localhost ~]$ sudo podman logs -f ipa-downloader
```
```
[core@localhost ~]$ sudo podman logs -f coreos-downloader
```
If the bootstrap VM cannot access the URL to the images, use the curl command to verify that the VM can access the images.
To inspect the bootkube logs that indicate if all the containers launched during the deployment phase, execute the following:
```
[core@localhost ~]$ journalctl -xe
```
```
[core@localhost ~]$ journalctl -b -f -u bootkube.service
```
Verify all the pods, including dnsmasq, mariadb, httpd, and ironic, are running:
```
[core@localhost ~]$ sudo podman ps
```
If there are issues with the pods, check the logs of the containers with issues. To check the log of the ironic-api, execute the following:
```
[core@localhost ~]$ sudo podman logs <ironic-api>
```

Cluster nodes will not PXE boot

When OKD cluster nodes will not PXE boot, execute the following checks on the cluster nodes that will not PXE boot. This procedure does not apply when installing a OKD cluster without the provisioning network.

Procedure

Check the network connectivity to the provisioning network.
Ensure PXE is enabled on the NIC for the provisioning network and PXE is disabled for all other NICs.

Verify that the install-config.yaml configuration file has the proper hardware profile and boot MAC address for the NIC connected to the provisioning network. For example:

control plane node settings

bootMACAddress: 24:6E:96:1B:96:90 # MAC of bootable provisioning NIC
hardwareProfile: default          #control plane node settings

Worker node settings

bootMACAddress: 24:6E:96:1B:96:90 # MAC of bootable provisioning NIC
hardwareProfile: unknown          #worker node settings

The API is not accessible

When the cluster is running and clients cannot access the API, domain name resolution issues might impede access to the API.

Procedure

Hostname Resolution: Check the cluster nodes to ensure they have a fully qualified domain name, and not just localhost.localdomain. For example:
```
$ hostname
```
If a hostname is not set, set the correct hostname. For example:
```
$ hostnamectl set-hostname <hostname>
```

Incorrect Name Resolution: Ensure that each node has the correct name resolution in the DNS server using dig and nslookup. For example:

$ dig api.<cluster-name>.example.com

; <<>> DiG 9.11.4-P2-RedHat-9.11.4-26.P2.el8 <<>> api.<cluster-name>.example.com
;; global options: +cmd
;; Got answer:
;; ->>HEADER<<- opcode: QUERY, status: NOERROR, id: 37551
;; flags: qr aa rd ra; QUERY: 1, ANSWER: 1, AUTHORITY: 1, ADDITIONAL: 2

;; OPT PSEUDOSECTION:
; EDNS: version: 0, flags:; udp: 4096
; COOKIE: 866929d2f8e8563582af23f05ec44203d313e50948d43f60 (good)
;; QUESTION SECTION:
;api.<cluster-name>.example.com. IN A

;; ANSWER SECTION:
api.<cluster-name>.example.com. 10800 IN	A 10.19.13.86

;; AUTHORITY SECTION:
<cluster-name>.example.com. 10800 IN NS	<cluster-name>.example.com.

;; ADDITIONAL SECTION:
<cluster-name>.example.com. 10800 IN A	10.19.14.247

;; Query time: 0 msec
;; SERVER: 10.19.14.247#53(10.19.14.247)
;; WHEN: Tue May 19 20:30:59 UTC 2020
;; MSG SIZE  rcvd: 140

The output in the foregoing example indicates that the appropriate IP address for the api.<cluster-name>.example.com VIP is 10.19.13.86. This IP address should reside on the baremetal network.

Cleaning up previous installations

In the event of a previous failed deployment, remove the artifacts from the failed attempt before attempting to deploy OKD again.

Procedure

Power off all bare metal nodes prior to installing the OKD cluster:

$ ipmitool -I lanplus -U <user> -P <password> -H <management-server-ip> power off

Remove all old bootstrap resources if any are left over from a previous deployment attempt:

for i in $(sudo virsh list | tail -n +3 | grep bootstrap | awk {'print $2'});
do
  sudo virsh destroy $i;
  sudo virsh undefine $i;
  sudo virsh vol-delete $i --pool $i;
  sudo virsh vol-delete $i.ign --pool $i;
  sudo virsh pool-destroy $i;
  sudo virsh pool-undefine $i;
done

Remove the following from the clusterconfigs directory to prevent Terraform from failing:

$ rm -rf ~/clusterconfigs/auth ~/clusterconfigs/terraform* ~/clusterconfigs/tls ~/clusterconfigs/metadata.json

Issues with creating the registry

When creating a disconnected registry, you might encounter a "User Not Authorized" error when attempting to mirror the registry. This error might occur if you fail to append the new authentication to the existing pull-secret.txt file.

Procedure

Check to ensure authentication is successful:

$ /usr/local/bin/oc adm release mirror \
  -a pull-secret-update.json
  --from=$UPSTREAM_REPO \
  --to-release-image=$LOCAL_REG/$LOCAL_REPO:${VERSION} \
  --to=$LOCAL_REG/$LOCAL_REPO

Example output of the variables used to mirror the install images:

UPSTREAM_REPO=${RELEASE_IMAGE}
LOCAL_REG=<registry_FQDN>:<registry_port>
LOCAL_REPO='ocp4/openshift4'

The values of RELEASE_IMAGE and VERSION were set during the Retrieving OpenShift Installer step of the Setting up the environment for an OpenShift installation section.

After mirroring the registry, confirm that you can access it in your disconnected environment:

$ curl -k -u <user>:<password> https://registry.example.com:<registry-port>/v2/_catalog
{"repositories":["<Repo-Name>"]}

Miscellaneous issues

Addressing the `runtime network not ready` error

After the deployment of a cluster you might receive the following error:

`runtime network not ready: NetworkReady=false reason:NetworkPluginNotReady message:Network plugin returns error: Missing CNI default network`

The Cluster Network Operator is responsible for deploying the networking components in response to a special object created by the installer. It runs very early in the installation process, after the control plane (master) nodes have come up, but before the bootstrap control plane has been torn down. It can be indicative of more subtle installer issues, such as long delays in bringing up control plane (master) nodes or issues with apiserver communication.

Procedure

Inspect the pods in the openshift-network-operator namespace:

$ oc get all -n openshift-network-operator

NAME                                    READY STATUS            RESTARTS   AGE
pod/network-operator-69dfd7b577-bg89v   0/1   ContainerCreating 0          149m

On the provisioner node, determine that the network configuration exists:

$ kubectl get network.config.openshift.io cluster -oyaml

apiVersion: config.openshift.io/v1
kind: Network
metadata:
  name: cluster
spec:
  serviceNetwork:
  - 172.30.0.0/16
  clusterNetwork:
  - cidr: 10.128.0.0/14
    hostPrefix: 23
  networkType: OpenShiftSDN

If it does not exist, the installer did not create it. To determine why the installer did not create it, execute the following:

$ openshift-install create manifests

Check that the network-operator is running:

$ kubectl -n openshift-network-operator get pods

Retrieve the logs:
```
$ kubectl -n openshift-network-operator logs -l "name=network-operator"
```
On high availability clusters with three or more control plane (master) nodes, the Operator will perform leader election and all other Operators will sleep. For additional details, see Troubleshooting.

Cluster nodes not getting the correct IPv6 address over DHCP

If the cluster nodes are not getting the correct IPv6 address over DHCP, check the following:

Ensure the reserved IPv6 addresses reside outside the DHCP range.

In the IP address reservation on the DHCP server, ensure the reservation specifies the correct DHCP Unique Identifier (DUID). For example:

# This is a dnsmasq dhcp reservation, 'id:00:03:00:01' is the client id and '18:db:f2:8c:d5:9f' is the MAC Address for the NIC
id:00:03:00:01:18:db:f2:8c:d5:9f,openshift-master-1,[2620:52:0:1302::6]

Ensure that route announcements are working.
Ensure that the DHCP server is listening on the required interfaces serving the IP address ranges.

Cluster nodes not getting the correct hostname over DHCP

During IPv6 deployment, cluster nodes must get their hostname over DHCP. Sometimes the NetworkManager does not assign the hostname immediately. A control plane (master) node might report an error such as:

Failed Units: 2
  NetworkManager-wait-online.service
  nodeip-configuration.service

This error indicates that the cluster node likely booted without first receiving a hostname from the DHCP server, which causes kubelet to boot with a localhost.localdomain hostname. To address the error, force the node to renew the hostname.

Procedure

Retrieve the hostname:
```
[core@master-X ~]$ hostname
```
If the hostname is localhost, proceed with the following steps.

Where X is the control plane node (also known as the master node) number.
Force the cluster node to renew the DHCP lease:
```
[core@master-X ~]$ sudo nmcli con up "<bare-metal-nic>"
```
Replace <bare-metal-nic> with the wired connection corresponding to the baremetal network.
Check hostname again:
```
[core@master-X ~]$ hostname
```
If the hostname is still localhost.localdomain, restart NetworkManager:
```
[core@master-X ~]$ sudo systemctl restart NetworkManager
```
If the hostname is still localhost.localdomain, wait a few minutes and check again. If the hostname remains localhost.localdomain, repeat the previous steps.
Restart the nodeip-configuration service:
```
[core@master-X ~]$ sudo systemctl restart nodeip-configuration.service
```
This service will reconfigure the kubelet service with the correct hostname references.
Reload the unit files definition since the kubelet changed in the previous step:
```
[core@master-X ~]$ sudo systemctl daemon-reload
```

Restart the kubelet service:

[core@master-X ~]$ sudo systemctl restart kubelet.service

Ensure kubelet booted with the correct hostname:

[core@master-X ~]$ sudo journalctl -fu kubelet.service

If the cluster node is not getting the correct hostname over DHCP after the cluster is up and running, such as during a reboot, the cluster will have a pending csr. Do not approve a csr, or other issues might arise.

Addressing a csr

Get CSRs on the cluster:
```
$ oc get csr
```

Verify if a pending csr contains Subject Name: localhost.localdomain:

$ oc get csr <pending_csr> -o jsonpath='{.spec.request}' | base64 --decode | openssl req -noout -text

Remove any csr that contains Subject Name: localhost.localdomain:
```
$ oc delete csr <wrong_csr>
```

Routes do not reach endpoints

During the installation process, it is possible to encounter a Virtual Router Redundancy Protocol (VRRP) conflict. This conflict might occur if a previously used OKD node that was once part of a cluster deployment using a specific cluster name is still running but not part of the current OKD cluster deployment using that same cluster name. For example, a cluster was deployed using the cluster name openshift, deploying three control plane (master) nodes and three worker nodes. Later, a separate install uses the same cluster name openshift, but this redeployment only installed three control plane (master) nodes, leaving the three worker nodes from a previous deployment in an ON state. This might cause a Virtual Router Identifier (VRID) conflict and a VRRP conflict.

Get the route:
```
$ oc get route oauth-openshift
```

Check the service endpoint:

$ oc get svc oauth-openshift

NAME              TYPE        CLUSTER-IP      EXTERNAL-IP   PORT(S)   AGE
oauth-openshift   ClusterIP   172.30.19.162   <none>        443/TCP   59m

Attempt to reach the service from a control plane (master) node:

[core@master0 ~]$ curl -k https://172.30.19.162

{
  "kind": "Status",
  "apiVersion": "v1",
  "metadata": {
  },
  "status": "Failure",
  "message": "forbidden: User \"system:anonymous\" cannot get path \"/\"",
  "reason": "Forbidden",
  "details": {
  },
  "code": 403

Identify the authentication-operator errors from the provisioner node:

$ oc logs deployment/authentication-operator -n openshift-authentication-operator

Event(v1.ObjectReference{Kind:"deployment", Namespace:"openshift-authentication-operator", Name:"authentication-operator", UID:"225c5bd5-b368-439b-9155-5fd3c0459d98", APIVersion:"apps/v1", ResourceVersion:"", FieldPath:""}): type: 'Normal' reason: 'OperatorStatusChanged' Status for clusteroperator/authentication changed: Degraded message changed from "IngressStateEndpointsDegraded: All 2 endpoints for oauth-server are reporting"

Solution

Ensure that the cluster name for every deployment is unique, ensuring no conflict.
Turn off all the rogue nodes which are not part of the cluster deployment that are using the same cluster name. Otherwise, the authentication pod of the OKD cluster might never start successfully.

Failed Ignition during Firstboot

During the Firstboot, the Ignition configuration may fail.

Procedure

Connect to the node where the Ignition configuration failed:
```
Failed Units: 1
  machine-config-daemon-firstboot.service
```

Restart the machine-config-daemon-firstboot service:

[core@worker-X ~]$ sudo systemctl restart machine-config-daemon-firstboot.service

NTP out of sync

The deployment of OKD clusters depends on NTP synchronized clocks among the cluster nodes. Without synchronized clocks, the deployment may fail due to clock drift if the time difference is greater than two seconds.

Procedure

Check for differences in the AGE of the cluster nodes. For example:

$ oc get nodes

NAME                         STATUS   ROLES    AGE   VERSION
master-0.cloud.example.com   Ready    master   145m   v1.16.2
master-1.cloud.example.com   Ready    master   135m   v1.16.2
master-2.cloud.example.com   Ready    master   145m   v1.16.2
worker-2.cloud.example.com   Ready    worker   100m   v1.16.2

Check for inconsistent timing delays due to clock drift. For example:

$ oc get bmh -n openshift-machine-api

master-1   error registering master-1  ipmi://<out-of-band-ip>

$ sudo timedatectl

               Local time: Tue 2020-03-10 18:20:02 UTC
           Universal time: Tue 2020-03-10 18:20:02 UTC
                 RTC time: Tue 2020-03-10 18:36:53
                Time zone: UTC (UTC, +0000)
System clock synchronized: no
              NTP service: active
          RTC in local TZ: no

Addressing clock drift in existing clusters

Create a Butane config file including the contents of the chrony.conf file to be delivered to the nodes. In the following example, create 99-master-chrony.bu to add the file to the control plane nodes. You can modify the file for worker nodes or repeat this procedure for the worker role.

See "Creating machine configs with Butane" for information about Butane.

variant: openshift
version: 4.8.0
metadata:
  name: 99-master-chrony
  labels:
    machineconfiguration.openshift.io/role: master
storage:
  files:
  - path: /etc/chrony.conf
    mode: 0644
    overwrite: true
    contents:
      inline: |
        server <NTP-server> iburst (1)
        stratumweight 0
        driftfile /var/lib/chrony/drift
        rtcsync
        makestep 10 3
        bindcmdaddress 127.0.0.1
        bindcmdaddress ::1
        keyfile /etc/chrony.keys
        commandkey 1
        generatecommandkey
        noclientlog
        logchange 0.5
        logdir /var/log/chrony

1	Replace `<NTP-server>` with the IP address of the NTP server.

Use Butane to generate a MachineConfig object file, 99-master-chrony.yaml, containing the configuration to be delivered to the nodes:
```
$ butane 99-master-chrony.bu -o 99-master-chrony.yaml
```
Apply the MachineConfig object file:
```
$ oc apply -f 99-master-chrony.yaml
```

Ensure the System clock synchronized value is yes:

$ sudo timedatectl

               Local time: Tue 2020-03-10 19:10:02 UTC
           Universal time: Tue 2020-03-10 19:10:02 UTC
                 RTC time: Tue 2020-03-10 19:36:53
                Time zone: UTC (UTC, +0000)
System clock synchronized: yes
              NTP service: active
          RTC in local TZ: no

To setup clock synchronization prior to deployment, generate the manifest files and add this file to the openshift directory. For example:

$ cp chrony-masters.yaml ~/clusterconfigs/openshift/99_masters-chrony-configuration.yaml

Then, continue to create the cluster.

Reviewing the installation

After installation, ensure the installer deployed the nodes and pods successfully.

Procedure

When the OKD cluster nodes are installed appropriately, the following Ready state is seen within the STATUS column:

$ oc get nodes

NAME                   STATUS   ROLES           AGE  VERSION
master-0.example.com   Ready    master,worker   4h   v1.16.2
master-1.example.com   Ready    master,worker   4h   v1.16.2
master-2.example.com   Ready    master,worker   4h   v1.16.2

Confirm the installer deployed all pods successfully. The following command removes any pods that are still running or have completed as part of the output.
```
$ oc get pods --all-namespaces | grep -iv running | grep -iv complete
```

Troubleshooting the installer workflow

Troubleshooting install-config.yaml

Bootstrap VM issues

Bootstrap VM cannot boot up the cluster nodes

Inspecting logs

Cluster nodes will not PXE boot

The API is not accessible

Cleaning up previous installations

Issues with creating the registry

Miscellaneous issues

Addressing the runtime network not ready error

Cluster nodes not getting the correct IPv6 address over DHCP

Cluster nodes not getting the correct hostname over DHCP

Routes do not reach endpoints

Failed Ignition during Firstboot

NTP out of sync

Reviewing the installation

Troubleshooting `install-config.yaml`

Addressing the `runtime network not ready` error