This is the fifth installment in the series of blog posts, “Last Month in Nakadi”, where I try to give some more details on new features and bug fixes that were released over the previous month.
[Changed] Reduced logging
Nakadi logs a lot of stuff. It’s very useful, but also comes with a cost. Recently, we were looking at our logs, and noticed that our SLO logging amounts for a large percentage of our logs. So we combined it with our access log, which significantly reduced the number of lines logged.
[Changed] Stricter JSON parser
Pull Request 1
Pull Request 2
While working on an internal service that consumes data from Nakadi, we noticed that the JSON parser in Nakadi is a bit too lax, and allows event producers to publish incorrect JSON. The issue is that our new service’s parser couldn’t parse that. In this change, we implemented a new parser, which is both more strict and more efficient. So, Nakadi performs a little bit better, and consumers have stronger guarantees regarding the events they consume.
In order to avoid breaking existing producers, we released this feature in two parts: first, we would log events that would be accepted by the old parser but not by the new one. We ran this version for a few days and got in touch with affected producers. Once we were satisfied that all producers would not be affected, we release the second pull request, which only uses the new, stricter parser.
[New] Feature flag to allow the deletion of event types with subscriptions
So far, deleting an event type has only been possible if there were no subscriptions for this event type. The reason behind this is to make sure that no consumers are taken by surprise when an event type is deleted. In our staging deployment, we found that it can sometimes cause issues and delays, especially when consuming applications are configured to automatically re-create subscriptions when they are deleted.
Now, there is a new feature flag, XXX, that allows users to delete event types that have subscriptions attached to them. Note that this will also delete the subscriptions, so we do not recommend turning this feature on for production systems – accidents happen!
[Removed] No more feature flag for subscriptions API
A few months ago, we announced that the low-level consumption API was now considered deprecated, and that clients should not use it anymore. We have not yet set a deadline for its removal from the code, but it will come eventually. To consume events from Nakadi, clients should now use the subscription API, also known as HiLA (High-Level API). This API could be turned on and off with a feature toggle, but since it is now the only supported one, the toggle doesn’t make much sense anymore. So we removed it, in order to make the code a little easier to read and maintain.
[New] New attributes
Pull Request 1
Pull Request 2
These two pull requests add optional attributes to an event type, for consistency with Zalando’s REST API guidelines.
The first one is
ordering_key_fields, which can be used when the order of events across multiple partitions is important. Nakadi only guarantees ordering within partitions, so this can be used by consumers to re-order events once they have been consumed.
The second one is
audience, which determines the target of an event type. The target is a class of consumers, such as
external_partner, or others.
Both attributes are optional, and Nakadi does not perform any kind of enforcement as a result of these being set – they are purely informational.
There was a bug in the subscription stats endpoint. In some rare cases, it could look like there were unconsumed events in a subscription, but that was not actually the case. No events would be lost, but monitoring systems would report unconsumed events – which couldn’t be consumed, since they didn’t actually exist.
And that’s it for June. If you would like to contribute to Nakadi, please feel free to browse the issues on github, especially those marked with the “help wanted” tag. If you would like to implement a large new feature, please open an issue first to discuss it, so we can all agree on what it should look like. We very much welcome all sorts of contributions: not just code, but also documentation, help with the website, etc.