Rolling Updates

Introduction

A rolling update is the process of replacing one version of the system with another without downtime. The changes can be new code, changed dependencies such as new Akka version, or modified configuration. Rolling updates are typically used for a stateful Akka Cluster where you can’t run two separate clusters in parallel during the update as in blue green deployments.

There are two parts of Akka that need careful consideration when performing an rolling update.

  1. Compatibility of remote message protocols. Old nodes may send messages to new nodes and vice versa.
  2. Serialization format of persisted events and snapshots. New nodes must be able to read old data, and during the update old nodes must be able to read data stored by new nodes.

There are many more application specific aspects. It’s important to have a strategy for serialization format that can be evolved and you can find advice in Persistence - Schema Evolution, which also applies to remote messages when deploying with rolling updates.

Akka supports rolling updates between two consecutive patch versions unless an exception is mentioned on this page. For example updating Akka version from 2.5.15 to 2.5.16. Many times it is also possible to skip several versions and exceptions to that are also described here. For example it’s possible to update from 2.5.14 to 2.5.16 without intermediate 2.5.15.

It’s not supported to have a cluster with more than two different versions. Roll out the first update completely before starting next update.

Shutdown order

It’s best to keep the node that has been running the longest until last. That is because Cluster Singletons (including Cluster Sharding coordinators) are running on the oldest nodes. It’s more efficient to avoid moving the singletons more than necessary since those typically have to recover their state and it might introduce unnecessary delays in for example access to new sharded actors.

Change log

2.5.0 Several changes in minor release

See migration guide when updating from 2.4.x to 2.5.x.

2.5.10 Joining regression

Issue: #24622

Incompatible change was introduced in 2.5.10 and fixed in 2.5.11.

This means that you can’t do a rolling update from 2.5.9 to 2.5.10 and must instead do update from 2.5.9 to 2.5.11.

2.5.10 Joining old versions

Issue: #25491

Incompatibility was introduced in in 2.5.10 and fixed in 2.5.15.

That means that you should do rolling update from 2.5.9 directly to 2.5.15 if you need to be able to join 2.5.9 nodes during the update phase.

2.5.14 Distributed Data serializer for ORSet[ActorRef]

Issue: #23703

Intentional change was done in 2.5.14.

This change required a two phase update where the data was duplicated to be compatible with both old and new nodes.

  • 2.5.13 - old format, before the change. Can communicate with intermediate format and with old format.
  • 2.5.14, 2.5.15, 2.5.16 - intermediate format. Can communicate with old format and with new format.
  • 2.5.17 - new format. Can communicate with intermediate format and with new format.

This means that you can’t update from 2.5.13 directly to 2.5.17. You must first update to one of the intermediate versions 2.5.14, 2.5.15, or 2.5.16.

Found an error in this documentation? The source code for this page can be found here. Please feel free to edit and contribute a pull request.