Automatically upgrade Vault clusters

Operators running Vault Enterprise with integrated storage can use automated upgrades to upgrade the Vault version currently running in a cluster automatically. There are a few different ways to make this upgrade happen, and control which versions are being upgraded to. With no additional configuration, Vault will check the version of Vault that each node in the cluster is running. If a blue/green style deployment is desired, Vault uses the version of your choosing, regardless of which version of Vault is currently running.

Configuration

A new key can be added to Vault's storage configuration stanza: autopilot_upgrade_version. The value for this key is a SemVer compatible version string of your choosing. When a version string is present, it will override the current version of Vault that is running to upgrade automatically.

Mechanics

Whether you choose to use Vault's built-in version or a version of your own, the mechanics for performing automatic upgrades remain the same.

When a Vault cluster is running and new nodes containing an updated Vault version join the cluster, the Autopilot subsystem within Vault will promote the new version nodes to voters when the number of nodes running the latest Vault version equals or exceeds the number of pre-existing nodes. Vault then demotes the previous version's nodes to non-voters. Finally, leadership transfers from the prior leader to a randomly selected node running the newest Vault version and waits for the user to remove the previous nodes from the cluster.

Below is a flowchart depicting Autopilot's automated upgrade state machine.

The status of the automated upgrade can be monitored by consulting the Autopilot state API endpoint.

Examples

Using vault's built-in version

This is the easiest method to perform an automated upgrade; no configuration is needed, and automated upgrades are enabled by default.

Using a blue/green style deploy

Specify something like: 1.11.0-release.1 for the autopilot_upgrade_version configuration key in your existing cluster. When you're ready to deploy a new set of nodes, specify 1.11.0-release.2 for the new nodes. Any time you need to deploy an updated set of nodes to the cluster, increment the final number.

It's important to note that Autopilot uses strict semantic versioning when parsing upgrade versions. This means that in the above example, -release.1 is considered the pre-release version in semantic versioning terminology. The pre-release version is denoted by a hyphen - after the patch version. This is semantically different from the build metadata, which is denoted by a plus sign + after either the patch version or the pre-release version. Build metadata is ignored when parsing versions, while the pre-release version is not. A table with some examples will help illustrate this.