Help us improve your experience.

Let us know what you think.

Do you have time for a two-minute survey?

 
 

Updating Contrail Networking using Zero Impact Upgrade Procedure in a Canonical Openstack Multi-model Deployment with Juju Charms

This document provides the steps needed to update a Contrail Networking deployment that is using Canonical Openstack Multi-model as it’s orchestration platform. The procedure utilizes Juju charms and provides a zero impact upgrade (ZIU) with minimal disruption to network operations.

Prerequisites

This document makes the following assumptions about your environment:

  • A Contrail Networking deployment using Canonical Openstack Multi-model as the orchestration platform is already operational.

  • Juju charms for Contrail services are active in your environment, and the Contrail Networking controller has access to the Juju jumphost and the Juju cluster.

When to Use This Procedure

This procedure is used to upgrade Contrail Networking when it is running in environments using Canonical Openstack Multi-model.

You can use this procedure to incrementally upgrade to the next main Contrail Networking release only. This procedure is not validated for upgrades between releases that are two or more releases apart.

The procedure in this document has been validated for the following Contrail Networking upgrade scenarios:

Table 1: Contrail Networking with Canonical Openstack Multi-model Validated Upgrade Scenarios

Starting Contrail Networking Release

Target Upgraded Contrail Networking Release

21.4.L2 Release

21.4.L3 Release

Recommendations

We strongly recommend performing the following tasks before starting this procedure:

  • Backup your current environment.

  • Enable huge pages for the kernel-mode vRouter.

    Huge pages in Contrail Networking are used primarily to allocate flow and bridge table memory within the vRouter. Huge pages for kernel-mode vRouters provide enough flow and bridge table memory to avoid compute node reboots to complete future Contrail Networking software upgrades.

    We recommend allotting 2GB of memory—either using the default 1024x2MB huge page size setting or the 2x1GB size setting—for huge pages. Other huge page size settings should only be set by expert users in specialized circumstances.

    A compute node reboot is required to initially enable huge pages. Future compute node upgrades can happen without reboots after huge pages are enabled. The 1024x2MB huge page setting is configured by default starting in Contrail Networking Release 2005, but is not active in any compute node until the compute node is rebooted to enable the setting.

    2GB and 1MB huge page size settings cannot be enabled simultaneously in environments using Juju.

    The huge page configurations can be changed by entering one of the following commands:

    • Enable 1024 2MB huge pages (Default): juju config contrail-agent kernel-hugepages-2m=1024

      Disable 2MB huge pages (empty value): juju config contrail-agent kernel-hugepages-2m=““

    • Enable 2 1GB huge pages: juju config contrail-agent kernel-hugepages-1g=2

      Disable 1GB huge pages (default. empty value): juju config contrail-agent kernel-hugepages-1g=““

      Note:

      1GB huge page settings can only be specified at initial deployment; you cannot modify the setting in active deployments. The 1GB huge page setting can also not be completely disabled after being activated on a compute node. Be sure that you want to use 1GB huge page settings on your compute node before enabling the setting.

Updating Contrail Networking in a Canonical Openstack Multi-model Deployment Using Juju Charms

Before updating Contrail Networking in a Canonical Openstack Multi-model Deployment, you are required to prepare the multi-model setup.

Preparing Multi-model Setup

You can initiate the creation of multi-model setup during Juju bootstrap and specify the multi-model name with --add-model parameter. Let the initial model be called as 'openstack'. Perform the following steps to prepare the multi-model setup:

  1. Add 'openstack' model for Kubernetes.

  2. Deploy Contrail-Openstack bundle for openstack model.

  3. Add 'kubernetes' model for Kubernetes.

  4. Deploy k8s bundle.

  5. Add offer for contrail-controller and keystone (if you are using keystone authorization for k8s).

    If easyrsa is implemented in Openstack model, then run the following command:

  6. Add relations to contrail-openstack.

    If easyrsa is implemented in Openstack model, then run the following commands:

Zero Impact Upgrade Procedure for Multi-model Deployment

To update Contrail Networking in an environment that is using Canonical Openstack Multi-model as the orchestration platform:

  1. From the Juju jumphost, enter the run-action command to place all control plane services—Contrail Controller, Contrail Analytics, and Contrail AnalyticsDB—into maintenance mode in preparation for the upgrade.
    Note:

    The --wait option is not required to complete this step, but is recommended to ensure this procedure completes without interfering with the procedures in the next step.

    Wait for all charms to move to the maintenance status. You can check the status of all charms by entering the juju status command.

  2. Upgrade all charms. See the Upgrading applications document from Juju.
  3. Update the image tags on both modals with new-tag in Juju. The update applies to Contrail Analytics, Contrail AnalyticsDB, Contrail Agent, Contrail Openstack and Contrail Controller services.

    The upgrade process starts. It automatically passes six stages.

  4. After updating the image tags, wait for all services to complete stage 5 of the ZIU upgrade process workflow. The wait time for this step varies by environment, but often takes 40 minutes.

    Enter the juju status command and review the Workload and Message field outputs to monitor progress. The update is complete when all services are in the maintenance state—the Workload field output is maintenance—and each individual service has completed stage 5 of the ZIU upgrade—illustrated by the ziu is in progress - stage/done = 5/5 output in the Message field.

    A sample output of an in-progress update that has not completed the image tag update process. The Message field illustrates that the ZIU processes have not completed stage 5 of the upgrade.

  5. Upgrade every Contrail agent on each individual compute node:

    Wait for each compute node and CSN node upgrade to finish. The wait time for this step varies by environment, but typically takes around 5 minutes to complete per node.

  6. If huge pages are not enabled for your vRouter, log into each individual compute node and reboot to complete the procedure.
    Note:

    A compute node reboot is required to initially enable huge pages. If huge pages have been configured in Juju without a compute node reboot, you can also use this reboot to enable huge pages. You can avoid rebooting the compute node during future software upgrades after this initial reboot.

    1024x2MB huge page support is configured by default starting in Contrail Networking Release 2005, which is also the first Contrail Networking release that supports huge pages. If you are upgrading to Release 2005 for the first time, a compute node reboot is always required because huge pages could not have been previously enabled.

    This reboot also enables the default 1024x2MB huge page configuration unless you change the huge page configuration in Release 2005 or later.

    This step can be skipped if huge pages are enabled.