Skip to main content

6 posts tagged with "specification"

View All Tags

Feature Observability Semantic Conventions

ยท 3 min read
Daniel Dyla
Senior Open Source Architect

As comic book aficionados say, with great power comes great responsibility, or great complexity as the case may be. Feature flags provide incredible value, and now with OpenFeature they're more accessible than ever, but the complexity they introduce should not be ignored. As your application grows, it becomes more and more difficult to understand how feature flag evaluations and changes impact your users and their experience with your application. The OpenTelemetry feature observability semantic convention special interest group (SIG) aims to combat this problem by defining a consistent set of rules and definitions for feature telemetry data. Keep reading to learn more about this effort and how you can be involved!

OpenFeature Web SDK v1

ยท 4 min read
Michael Beemer
Product Manager at Dynatrace, OpenFeature maintainer

We are excited to announce the availability of the OpenFeature Web SDK v1. The Web SDK represents a culmination of years of research and industry consensus. It was built from the ground up to be performant, powerful, and flexible while providing broad support for your favorite feature management tool or in-house solution.

Reconciling with State: Simplifications and Improvements to Better Support Framework-Specific SDKs

ยท 5 min read
Todd Baert
Software Engineer at Dynatrace, OpenFeature maintainer

Recent Specification Changes ๐Ÿ—’โ€‹

There's no shortage of frameworks available when it comes to the development of enterprise software. Whether it's the "frontend" or "backend", developers are quick to avail things like React, Spring, Gin and Flask to avoid boilerplate code and structure their applications in a familiar way. Such frameworks also offer convenient abstractions that ease the employment of common patterns, think dependency injection as in Angular or Spring, or MVC as in Django or ASP.NET MVC. Many also provide extensibility features so that new functionality can be implemented seamlessly and idiomatically.

OpenFeature, we've been working on enhancing our specification and base SDKs to support the creation of such framework-level SDKs in order to bring vendor neutral feature flags to them. Specifically with respect to front end frameworks such as React, we've found that it was critical to refine the semantics of our events API and context-reconciliation concepts. This post discusses some of our improvements.

Context Reconciliationโ€‹

As discussed in previous blog posts, we've seen that client-side feature flag APIs share some commonalities we refer to collectively as the static-context paradigm; in short, the context used as the basis for flag evaluation is less frequently changed and generally corresponds to a user, or application state that a user occasionally modifies. When a user logs in, or some state is modified, providers frequently need to fetch a new ruleset or re-evaluate their cache of flags based on the new contextual data. We refer to this as reconciliation. During reconciliation, some asynchronous work is often required (frequently this takes form of some kind of network request). During this time, depending on the provider and the needs of the application, it might be necessary to display loading indicators, or prevent users from taking action. Some frameworks have patterns or in-built functionality to support this, such as React's suspense. In order to leverage such features, we've added a new RECONCILING provider state and PROVIDER_RECONCILING provider event. These indicate the provider is in the process of reconciling it's internal state due to a context change. Previously, something like this was achievable with the use of the STALE state and events, but we believed this was overloading the meaning of this state and didn't provide adequately robust semantics.

* transitions occurring when associated events are spontaneously emitted from the provider

Wider Implications: "Stateless" Providersโ€‹

Up to now, providers maintained a publicly-exposed representation of their own state: when they became ready, they changed their state to READY; if they lost connection to their source of truth, they might change their state to STALE or ERROR and emit PROVIDER_STALE or PROVIDER_ERROR events. This state management was burdensome for provider authors, and the addition of new states (such as RECONCILING mentioned above) only worsened that issue. Additionally, it allowed for widely divergent behavior between provider implementations, making the creation of framework-specific SDKs difficult and their behavior less predictable between providers, without meaningful benefit. With this update, the provider interface has become stateless. Now, the SDK itself maintains the provider's state based on the completion of lifecycle functions (initialization, shutdown, on context change) and events emitted by the provider. This simplifies things for provider authors, and results in a uniform experience for application authors, especially when leveraging framework-specific SDKs.

Bad Statesโ€‹

In some circumstances, providers may not be able to evaluate flags. This could be because a provider hasn't yet been fully initialized, or has experienced some internal error. Previously, provider-authors had to ensure their publicly exposed state was updated if such errors occurred, as well as emit an error event. Furthermore, provider authors had to check with this state to accurately return an informative error code (ie: PROVIDER_NOT_READY) if evaluation was attempted in such states. With these recent changes, we've added a new FATAL error code, indicating that the provider has entered a irrecoverable error state (typically due to a bad credential, hostname or other value that's likely to represent a misconfiguration). Along with this helpful new state, implementations will now "short-circuit" evaluations if the provider is in NOT_READY or FATAL states.

Overall, we believe these changes will improve consistency for application authors across providers, and simultaneously reduce the burden on provider authors as well.

Contributing ๐Ÿคโ€‹

If you're interested in contributing, there's no shortage of opportunities! You can create a client-side provider for a vendor or tool you're familiar with and host in our contrib repos. If the server-side is more your thing, you can help to update any of our existing server-side providers by simplifying them consistent with this latest spec change. As always, we're looking for help with our newest SDKs, some of which have yet to reach 1.0, including Kotlin, Swift, Python, Ruby, and Rust. We're also interested in more framework-specific SDKs as discussed in this earlier post. If there's a framework or library you think could benefit from an OpenFeature SDK, these are a great idea for a first contribution.

Shout Outs ๐Ÿฅ‡โ€‹

Thanks to everyone who worked on this specification change, and a special shout-out to the folks at Spotify who've been contributing to the spec, as well as to our client implementations including our Kotlin and Swift SDKs.

Stay Tuned... ๐Ÿ“ปโ€‹

We're excited to be working on an open, vendor-neutral protocol for the evaluation of feature flags, which adheres to OpenFeature semantics. This protocol could be used as the basis for a single provider which is compatible with any implementing backend, and might be of particular usefulness for client-side feature flagging. If you'd like to know more, check out our dedicated slack channel for this topic.

Specification 0.6.0

ยท 3 min read
Michael Beemer
Product Manager at Dynatrace, OpenFeature maintainer

We are excited to announce the availability of version 0.6.0 of the OpenFeature spec. This release brings many significant additions that benefit both provider developers and application authors, while also establishing a strong foundation for first-class client-side support in OpenFeature.

Catering to the client-side

ยท 12 min read
Pete Hodgson
Software delivery consultant, OpenFeature contributor

While OpenFeature initially focused on support for server-side feature flagging, we know that a lot of feature-flagging (likely the majority) happens on the client - mobile apps and frontend web apps. As such, we're currently finalizing a proposal which extends the OpenFeature spec to support client-side use cases. By the way, if you're working on a feature flagging framework, whether it's commercial, open-source, or internal product, the folks at OpenFeature would love to hear more about how you approach client-side flagging.

In this post I'll summarize those changes, but to understand them in context we'll first talk about what makes client-side feature flagging different before diving into how that will impact the OpenFeature APIs.

Specification hardening and 1.0 SDKs

ยท 2 min read
Todd Baert
Software Engineer at Dynatrace, OpenFeature maintainer

Early this year, OpenFeature announced its intent to bring a standard to the rapidly growing development practice of feature flagging. In June it was accepted as a Cloud Native Computing Foundation Sandbox project. Now, we're pleased to announce a new milestone: OpenFeature has released 1.0 versions of its .NET, Go, Java, and JavaScript SDKs!