Today we are talking about two releases for our PHP library patchlevel/event-sourcing: version 3.9.0 and version 3.10.0!
The first release, 3.9.0, was a smaller one mostly focused on some DX improvements. The second, 3.10.0, comes with
some neat features for the subscription engine along with various updates and deprecations. So, let's start going
through these changes.
We made a change in patchlevel/worker which makes using the worker even more valuable. You can now utilize the events from the worker by adding your own listeners to its internal events. Previously, we already used an event dispatcher internally, but you could not add any listeners to it.
Now this is possible, which can come in handy if, for example, you are using a long-running process, like the subscription engine, and have an error tracking tool such as Sentry in place. Without these changes, it was not possible to manually flush messages to send them to Sentry if something happened during the execution of the subscription engine.
In the bundle, we added a ResetServicesListener which already
takes advantage of this new possibility. Now, after every run of the subscription engine, we are executing Symfony's
ServiceResetter. This makes the subscription engine even more resource-efficient than before. Also, regarding the
Sentry example, if you are using the Monolog variant, this now automatically flushes all errors to Sentry.
We streamlined the usage of ChildAggregates based on
a user request. The issue was that the ChildAggregate was a
private property of the aggregate and was implemented using our BasicAggregateRoot abstract class. This combination
led to an error due to visibility issues, which we have resolved by using reflection.
Keep in mind that the ChildAggregate feature is still experimental and that we also have another feature which we
believe is better suited in almost every
case: micro aggregates. You should probably
use that feature instead of ChildAggregates.
We officially added PHP 8.4 support, and not much work was needed. PHP 8.4 is now included in our test matrix; therefore, everything is tested with this new version.
We deprecated the AggregateIdArgumentResolver because we believe that if you need it, you have already done something
wrong. The Aggregate ID should be part of your event itself and not only part of our metadata. That's why we want to
remove this class from our library, as it may lead to poor event design.
We already supported defining a RetryStrategy for the subscription engine to handle failed attempts. Now, we are
taking it a step further by enabling you to define different
strategies for your subscribers. This gives you more control over the behavior when something goes wrong.
To make this work, we now have a RetryStrategyRepository which holds a map of unique name to strategy. This name
is then used for configuration.
To configure a retry strategy for a
subscriber, we offer a new attribute #[RetryStrategy] that expects the name of the strategy. The library comes with
two strategies configured out of the box: default and no_retry. The default strategy utilizes the
ClockBasedRetryStrategy, configured with 5 attempts, a base delay of 5 seconds, and a factor of 2. The
no_retry strategy is self-explanatory — it simply does not retry.
In version 3.7.0, we added a new status for subscriptions: the failed status. This status is reached when an error
occurs and the retry strategy fails to recover from it. Now, we have polished this further by allowing you to easily
hook into this case. We added a #[OnFailed] attribute that you can add to a method on the subscriber
to handle the failed message. For example, you
can add this message to a DeadLetter queue for later processing.
By doing this, you free the subscription engine from holding the subscription in the failed state. After successfully
handling the #[OnFailed] method, the subscription status will move from failed to active. When using this with
projections, be aware that data may then be missing from the table.
These two releases deliver some neat features, especially around the subscription engine, making it even more efficient and powerful. As always, we would love to hear your feedback about these features or any other topics related to event sourcing, so don't hesitate to open an issue on GitHub or start a discussion there. The next release is already in the pipeline, so stay tuned.
We are excited to announce the release of patchlevel/event-sourcing version 3.8.0! This update introduces features such as lookup, command bus, read-only event store, and the official release of the stream store. In this post, we will walk you through the key changes.
This article addresses the common concern regarding the performance of event sourcing, particularly the speed at which long-living aggregates with many events are loaded. It explores solutions such as snapshotting and stream splitting to optimize aggregate loading. Furthermore, projections allow for the creation of highly flexible and optimized read models, each can be tailored to specific needs.
We’re excited to announce the release of our php library patchlevel/event-sourcing version 3.7.0. This release features better testing capabilities with InMemorySubscriptionStore::clear, improved subscription performance and a new #[Stream] attribute for micro aggregates!