Operating and troubleshooting Autopilot in GCP Anthos

Summary and Key concepts

Summary:

This article provides troubleshooting procedures for monitoring objects managed by Portworx Autopilot. It explains how to view the status of objects through autopilotruleobject instances, which are created when a monitored object triggers a condition. Commands are provided to list and describe these rule objects, showing recent statuses and transitions. Additionally, the article covers how to generate a support bundle by signaling the Autopilot process, copying diagnostic files from the cluster, and capturing logs, which can then be sent to Portworx support for further assistance.

Kubernetes Concepts:

• Custom Resource Definitions (CRD): autopilotruleobject is a custom resource created when an Autopilot rule is triggered, storing the monitored object’s status and actions.
• kubectl exec: Executes commands in a running pod, such as sending a signal to the Autopilot process to create diagnostic files.
• kubectl get: Lists Kubernetes objects like autopilotruleobjects.

Portworx Concepts:

• Autopilot: Automates storage management tasks based on conditions and actions, like resizing volumes or rebalancing storage pools.

This section provides common operational procedures for monitoring and troubleshooting your autopilot installation.

Troubleshooting objects monitored by Autopilot

Follow the steps in the sections below to troubleshoot objects monitored by Autopilot

Get recent statuses using AutopilotRuleObjects

For each object monitored by Autopilot, it will create a corresponding autopilotruleobject instance in the namespace of the object.

• For volumes (PVCs), the autopilotruleobject instance will be in the namespace of the PVC.
• For storage pools, the autopilotruleobject instance will be in the namespace where Portworx is installed.

The autopilotruleobject is created only if an object's condiitons were atleast triggered once. In other words, you will not see an autopilotruleobject if the object was always in nornal state.

List all autopilotruleobjects

The following command lists all Autopilot rule objects in all namespaces:

kubectl get autopilotruleobjects --all-namespaces

Instead of entering the full autopilotruleobjects string, you can use the aro alias.

kubectl get aro --all-namespaces

NAMESPACE   NAME                                       AGE
pg1         pvc-xxxxxxxx-xxxx-xxxx-xxxx-03263da7b04c   1s

Describe a specific object

The Status section contains a list of recent object statuses:

kubectl describe aro -n pg1 pvc-xxxxxxxx-xxxx-xxxx-xxxx-03263da7b04c

Name:         pvc-xxxxxxxx-xxxx-xxxx-xxxx-03263da7b04c
Namespace:    pg1
Labels:       rule=volume-resize
Annotations:  <none>
API Version:  autopilot.libopenstorage.org/v1alpha1
Kind:         AutopilotRuleObject
Metadata:
  Creation Timestamp:  2020-08-26T22:29:45Z
  Generation:          2
  Owner References:
    API Version:           autopilot.libopenstorage.org/v1alpha1
    Block Owner Deletion:  true
    Controller:            true
    Kind:                  AutopilotRule
    Name:                  volume-resize
    UID:                   xxxxxxxx-xxxx-xxxx-xxxx-62fbd2d5dbbc
  Resource Version:        7554069
  Self Link:               /apis/autopilot.libopenstorage.org/v1alpha1/namespaces/pg1/autopilotruleobjects/pvc-xxxxxxxx-xxxx-xxxx-xxxx-03263da7b04c
  UID:                     xxxxxxxx-xxxx-xxxx-xxxx-37fe57310baf
Status:
  Items:
    Last Process Timestamp:  2020-08-26T22:29:45Z
    Message:                 rule: volume-resize:pvc-xxxxxxxx-xxxx-xxxx-xxxx-03263da7b04c transition from Normal => Triggered
    State:                   Triggered
    Last Process Timestamp:  2020-08-26T22:30:19Z
    Message:                 rule: volume-resize:pvc-xxxxxxxx-xxxx-xxxx-xxxx-03263da7b04c transition from Triggered => ActionAwaitingApproval
    State:                   ActionAwaitingApproval
Events:                      <none>

List autopilotruleobjects for a given autopilotrule

You can use the label selector rule=<RULE_NAME> for list autopilotruleobjects only for that autopilotrule.

kubectl get aro --all-namespaces -l rule=volume-resize

NAMESPACE   NAME                                       AGE
pg1         pvc-xxxxxxxx-xxxx-xxxx-xxxx-03263da7b04c   4m3s

Troubleshooting autopilot

1. Create a directory (ap-cores) in which to store your support bundle files and send the support signal to the autopilot process:

   mkdir ap-cores
   POD=$(kubectl get pods -n portworx -l name=autopilot | grep -v NAME | awk '{print $1}')
   kubectl exec -n portworx $POD -- killall -SIGUSR1 autopilot

2. Copy the support bundle files from your Kubernetes cluster to your directory:

   kubectl cp  portworx/$POD:/tmp/aut-diags.zip ap-cores/aut-diags.zip
   ls ap-cores

3. Collect and place your autopilot pod logs into an autopilot-pod.log file within your temporary directory:

   kubectl logs $POD -n portworx --tail=99999 > ap-cores/autopilot-pod.log

Once you've created a support bundle and collected your logs, send all of the files in the ap-cores/ directory to Portworx support.