Upgrading PingDirectory ¶
Because PingDirectory is essentially a database, in its container form, each node in a cluster has its own persisted volume. Additionally, because PingDirectory is an application that focuses on state, rolling out an upgrade isn't really the same as any other configuration update. However, the product software, and scripts in the image provide a process through which upgrades are drastically simplified.
This use case focuses on a PingDirectory upgrade in a default Kubernetes environment where you upgrade a PingDirectory StatefulSet 9.0.0.1 to 9.1.0.0 using an incremental canary roll-out.
Tips ¶
To ensure a successful upgrade process:
- Avoid combining configuration changes and version upgrades in the same rollout. This adds unnecessary complexity to debugging errors during an upgrade.
- Successfully complete an upgrade in a proper Dev/QA environment before trying anything in production.
- Upgrades will happen on one server at a time. Ensure you have enough resources on the remaining machines to prevent client impact.
- Follow a canary deployment pattern to ensure the changes will be successful before doing a full rollout. There is no good way to roll back a completed upgrade, so take any necessary steps to avoid this.
Before you begin ¶
You must:
- Complete Get Started to set up your DevOps environment and run a test deployment of the products using Helm.
- Clone or download the
pingidentity-devops-getting-started/30-helm/pingdirectory-upgrade-partition
repository to your local${HOME}/projects/devops
directory. - Understand how to use our DevOps server profiles.
- Have access to a Kubernetes cluster.
- If you're upgrading in your own environment and using mounted licenses, have the license for the existing version in the
/opt/out
persisted volume and a license for the new version needs to be in/opt/in
.
The license locations aren't an needed if you're using our DevOps credentials in an evaluation context.
About this task ¶
You will:
- Start with a base stack.
- Set up a partition to make changes on just one node.
- Deploy changes to one node and fix any errors.
- Rollout changes to other nodes.
Upgrade process overview ¶
The key functionality for PingDirectory upgrades is the relationship between the image hooks and the manage-profile
command in the product.
The upgrade is processed as follows:
-
When a node starts for the first time, it' i's in
SETUP
mode and runsmanage-profile setup
. -
When a node restarts (for whatever reason), it runs
manage-profile replace-profile
.-
This command compares the new profile to the old profile, and if there is a change, it tries standing up the server with the new profile.
-
Errors are thrown if there's a configuration in the new profile that prevents it from being applied.
-
-
If
manage-profile replace-profile
detects a product version difference, it takes the same approach as any other restart. It attempts to migrate PingDirectory to the new version, and if it can't, the command fails.Because there is processing that happens automatically in the container during the upgrade, you should roll the change out to a small partition first and test it thoroughly before rolling it out to all. This partition also gives us room to revert back without impacting traffic in case something is not as expected.
This process in standard Kubernetes is called a Canary Rollout and is derived from Kubernetes documentation on StatefulSet update strategies.
"Canary Rollout" in this scenario focuses only on an incremental rollout of containers, not actually separating traffic. That could be done with additional tools like Istio or standing up another service and pairing separate labels and selectors.
Setting up the base stack ¶
The YAML configuration files for this use case are in your cloned local copy of the pingidentity-devops-getting-started
repository
-
To use the
1-initial.yaml
file in your localpingidentity-devops-getting-started/30-helm/pingdirectory-upgrade-partition
directory to start with a PingDirectory StatefulSet using persistent volumes, enter:helm upgrade --install releasename pingidentity/ping-devops -f 1-initial.yaml
All helm commands for this use case need to be run from the
pingidentity-devops-getting-started/30-helm/pingdirectory-upgrade-partition
directory.This stands up a two directory topology, each with its own Persistent Volume Claim using the default storage class.
-
Wait for both nodes to be healthy before continuing and then enter:
kubectl get pods
pingdirectory-0
andpingdirectory-1
should show1/1
in theREADY
column.
Setting up a partition ¶
To use the 2-partition.yaml
file in your local pingidentity-devops-getting-started/30-helm/pingdirectory-upgrade-partition
directory to add a partition to StatefulSet
for updateStrategy
, enter:
helm upgrade --install releasename pingidentity/ping-devops -f 2-partition.yaml
This partition configuration signifies that any changes to spec.template
will only be applied to nodes with a cardinal value higher than the partition value.
The only other change is to the image tag. When this change is applied:
-
The
pingdirectory-1
pod is terminated and a new one with the new image is started. -
The new PingDirectory container is based on a specific version of PingDirectory, found in the
/opt/server
volume. -
The
manage-profile replace-profile
command is eventually triggered when the container detects PingDirectory is in aRESTART
state. This command identifies the difference in the database version running, based on the persisted volume attached to/opt/out
, and then attempts to upgrade. Information similar to the following will be displayed:... pingdirectory-1 Validating source and existing servers ..... Done pingdirectory-1 Updating the server version from 8.0.0.1 to 8.1.0.0. Local database backends pingdirectory-1 will be exported before the update in case a revert is necessary pingdirectory-1 Exporting backend with backendID userRoot. This may take a while ..... Done pingdirectory-1 Running the update tool ..... Done ... pingdirectory-1 Cleaning up after replace ..... Done pingdirectory-1 manage-profile replace-profile returned 0
This process requires having licenses for both server versions available and in the right location.
If
manage-profile replace-profile
completes without error, you'll see the container continue the migration and eventually start up PingDirectory.If
manage-profile replace-profile
fails, an error is displayed and the container exits. The errors are because of some conflict in the server profile. The partition you set up previously provides an isolated environment for working out errors.Work through any errors until you can get
manage-profile replace-profile
to complete successfully before continuing.
Rolling out the changes ¶
When you're confident your upgrade will occur smoothly:
To use the 3-rollout-full-upgrade.yaml
file in your local pingidentity-devops-getting-started/30-helm/pingdirectory-upgrade-partition
directory to deploy the rollout to the remaining nodes, enter:
helm upgrade --install releasename pingidentity/ping-devops -f 3-rollout-full-upgrade.yaml
This removes the partition.