The past few months have seen a continued interest in Linkerd's data plane “micro-proxy”, Linkerd2-proxy. Last month, William Morgan wrote about some of the decisions that first went into building a service-mesh-specific proxy in Rust. Eliza Weisman then followed up with a deep dive into Linkerd2-proxy and how it handles requests. In this article, I want to look into the future: what do the upcoming months and years have in store for Linkerd's proxy? And how can you get involved?
Linkerd's secret sauce: a small, speedy, simple micro-proxy
Building a modern, high-performance Rust network proxy has not been a trivial undertaking. Especially in the early years, investing in the proxy often meant spending more time working on the core Rust networking libraries than on the proxy itself! But we knew that this would be critical to our service mesh approach: because the data plane scales directly with the application, a service mesh can only be small and fast if the underlying data plane itself is small and fast.
Happily, those early investments have paid off. As William wrote:
I believe that Linkerd2-proxy represents the state of the art for secure, modern network programming. It is fully asynchronous and written in a modern type-safe and memory-safe language. It makes full use of the modern Rust networking ecosystem, sharing foundations with projects such as Amazon’s Firecracker. It has native support for modern network protocols such as gRPC, can load balance requests based on real-time latency, and do protocol detection for zero-config use. It is fully open source, audited, and widely tested at scale.
Of course, Linkerd isn't just known for performance and resource usage, it's known for its simplicity. Here, again, Linkerd2-proxy is critical to Linkerd's approach: because Linkerd2-proxy is designed explicitly for the sidecar service mesh use case, it's dramatically simpler to operate than general-purpose proxies such as NGINX and Envoy. This is not a knock on those proxies; it's simply a reflection that by shedding all the non-service-mesh use cases, Linkerd2-proxy gains the leeway to “just work” without tuning or tweaking, through features like protocol detection and Kubernetes-native service discovery. In fact, at this point Linkerd-proxy doesn't even have a config file.
The future of the Linkerd2-proxy
Today, Linkerd-proxy is fast, simple, and small, and powers the critical production architecture of organizations around the world. So what's next?
There are a couple notable exciting future features on the docket, ranging from the concrete and near-term to the speculative and far out, and from the small and well-contained to the big and hairy. These include:
- mTLS for all TCP connections. This is a big one. Today, the proxy can automatically initiate and terminate mutually-authenticated TLS connections without any configuration, but only for HTTP/gRPC traffic. We're already working on extend this to non-HTTP protocols, so that they have the same guarantees of workload identity and confidentiality that Linkerd provides for HTTP traffic today. As an added bonus, this feature will also extend Linkerd's seamless multi-cluster capabilities to non-HTTP traffic.
- Revisiting latency bucketing. As part of its instrumentation, Linkerd2-proxy records the latency of all traffic that passes through it and reports these values in a set of fixed buckets, with a specific latency range (e.g. 5ms-10ms) per bucket. These bucket windows are fixed and frankly somewhat arbitrarily-chosen at the moment. We'd like to improve this, but any kind of dynamic sizing would probably require coordination between proxies so that the resulting buckets can be aggregated.
- HTTP/3 / QUIC between proxies. HTTP/2 provides some major benefits over HTTP/1.x for high-concurrency communication between services, and today, Linkerd2-proxy already transparently upgrades HTTP/1.1 communication to HTTP/2 when it happens in between proxies. HTTP/3 promises further benefits in this direction, and we're eager to extend Linkerd2-proxy to support HTTP/3.
- Web assembly plugins. Today, there's no immediate way of supporting plugins in the data plane, e.g. to allow for customized routing based on user-specific headers. We’re exploring proxy WebAssembly (Wasm) as a way to accomplish this and allow Linkerd to make use of a broad range of language-agnostic traffic plugins. There may even be a way to use the same plugins developed for Envoy, which could open up some very interesting possibilities.
- Off-cluster support. While the proxy can run outside of Kubernetes (it's just a binary, after all), there is no packaging today that makes this easy to do, and there several open questions about how things like mTLS identity should be treated. Nonetheless, it has come to our attention that apparently not all software in the universe runs on Kubernetes, and we'd like to better support that in Linkerd.
Some of those features, like mTLS for TCP, are well-scoped and work is already underway. Others, like HTTP/3, are still in idea phase. But they're all strategic and important for the proxy.
How can I get involved?
If any of those features sound interesting, or if the thought of working on a state-of-the-art open source Rust networking project that powers production systems around the world is exciting, then the good news is that you can get involved! Linkerd2-proxy is fully open source, open governance, and hosted by the Cloud Native Computing Foundation, and we're always interested in welcoming more folks to the project.
If you're new to Linkerd, the best way to get started is:
- Join the Linkerd slack, especially the
#contributorschannel, and say hi!
- Attend the next monthly Linkerd Online Community Meetup;
- Peruse the open proxy issues, especially those marked “good first issue”; and
- Familiarize yourself with our RFC process for introducing bigger changes to the project.
If you're new to Rust especially, you might also want to take a look at the live proxy code walkthroughs that Eliza has been running.
All the fame and glory of open source Rust maintainership awaits you. Join us and let's build this amazing proxy together.
Try it today!
Want to see the proxy in action? You can try Linkerd on any relatively modern Kubernetes cluster in a matter of minutes. Download the latest stable release by running:
curl https://run.linkerd.io/install | sh
You can be up and running with a functioning service mesh (and an awesome Rust proxy) in a matter of minutes.
Linkerd is for everyone
Linkerd is a community project and is hosted by the Cloud Native Computing Foundation. Linkerd is committed to open governance. If you have feature requests, questions, or comments, we'd love to have you join our rapidly-growing community! Linkerd is hosted on GitHub, and we have a thriving community on Slack, Twitter, and the mailing lists. Come and join the fun!