Towards a Sustainable Service Mesh

Open source covers a vast spectrum of projects, but they all face the same existential threat: what ensures this project will be maintained in the future? Today, Linkerd adopters can be confident they’re getting the world’s simplest, fastest, lightest service mesh. Can they be confident they’re getting a project that will be around tomorrow? Another decade?

These are not theoretical questions. The past 18 months alone have provided many examples of just how fragile open source can be:

In February, Weaveworks, the company behind much of the Gitops movement, shut its doors, temporarily leaving the fate of the widely-used CNCF-graduated Flux project in doubt.
In March, Redis, a well-established infrastructure project, changed its licenses to move away from open source, following a similar license change by Terraform a year earlier.
Last year, Open Service Mesh, a CNCF service mesh project created, maintained, and marketed heavily by Microsoft, shut down entirely, its maintainers reassigned to other projects.

Being a CNCF project—even a graduated one—offers no protection. The CNCF does not step in to rescue dying projects. Nor does the backing of a trillion-dollar company; nor a happy and robust community. Each project must find its own answer to this question.

To this end, in February we announced a significant change to Linkerd. We would no longer publish stable release packages as part of the open source project. Instead, we would rely on the vendor ecosystem for that work. The change itself was a small simplification of the project, but the goal behind it was very large: to ensure Linkerd could become a truly sustainable project, without relicensing, without violating CNCF rules, and without changing Linkerd’s fundamental open source nature.

We made this change the only way we thought was truly realistic: by building a clear economic path from the companies that build their businesses on top of Linkerd, to the vendors who employ, today, 100% of the Linkerd maintainers.

Did it work?

The Linkerd community is amazing

Yes. Today, we can confidently state that the future of Linkerd is very bright. You can read Buoyant’s full announcement here.

We first need to start with a giant thank you to the Linkerd community. Since our announcement, we’ve had countless conversations with adopters who understood the criticality of what we were trying to do and why it made sense for them; who told us they had our backs; who took this leap of faith with us.

Since making this change, the Linkerd team has gained both maintainers and momentum, achieving milestone after milestone:

We’ve shipped many features that have been languishing in the backlog, including support for IPv6, authorization policy audit mode, and bringing our Gateway API implementation to feature parity with ServiceProfiles.
We’ve delivered a new implementation of retries, timeouts, and per-route metrics—some of the earliest code ever shipped in Linkerd—that fixes some long-standing corner case behavior in these features.
We’ve made significant progress on both egress functionality, rate limiting, and Open Telemetry support, all long-awaited features to be delivered in the upcoming Linkerd 2.17 release.
We’ve added a search bar to linkerd.io docs. (Sometimes it’s the little things!)
We’ve hit 10,000 community Slack members :)

We’ve also significantly improved our edge release process to make edge artifacts easier to consume, providing both post-hoc guidance about each release as well as monthly roundups. As we hoped, this is has dramatically increased the speed of bugfixes for adopter-reported issues (see e.g. #12610).

Linkerd gets more maintainers

We’re very happy to report that Buoyant is adding maintainers to Linkerd to further increase the pace of development.

Two new maintainers (technically, still maintainers-in-training) have already joined the team. They are both increasing project bandwidth where we need it most: in Linkerd’s Rust microproxy, the nuclear engine behind Linkerd’s entire feature set. The proxy is the key to Linkerd’s stellar performance and simplicity, but it’s also the most challenging part of the project to work on. Linkerd’s microproxy is one of the most advanced Rust codebases in the world, and being able to operate effectively here requires an extremely high caliber of talent. These incredible folks will be working full time on making Linkerd’s dataplane even faster, even lighter, and even more featureful.

And that’s just the beginning.

Our advice to other CNCF projects

Shortly after our announcement, multiple CNCF maintainers reached out to us privately, wondering if a similar change might make sense for them.

Our answer today is “probably”. While open source projects can have dramatically different needs, constraints, and communities, many CNCF projects have a set of characteristics that we believe makes this type of release-artifact-focused change effective:

The project is primarily used in commercial contexts. Many CNCF projects, like Linkerd, have a community that is primarily there on behalf of their employer, on their employer’s dime, with the goal of advancing their employer’s business interests.
Adopters primarily interact with the project through release packages. Many CNCF projects, like Linkerd, have a community that primarily interacts with the projects by downloading and deploying pre-built release artifacts, and when the project is working they move on to the next task at hand.
The project is a critical part of adopters’ infrastructure. Many CNCF projects, like Linkerd, provide features which are critical to the business and cannot be easily replicated.

For projects like these, providing a clear way for adopters to fund project development is a tremendous asset. Our change gave Linkerd’s adopters agency in the project’s long-term survival, and the ability to not just de-risk the worst outcome but to accelerate the best ones. These are transformative benefits.

Our experience was also that the CNCF itself was receptive to this change. We met with the TOC shortly after the announcement. They asked for minor clarifications and adjustments to our plan, but the tone of the meeting was supportive, and it was clear that ultimately everyone, from TOC to maintainers to adopters, shared the same goal: the long-term health of Linkerd.

What’s next for Linkerd?

More, faster, better Linkerd-ing, of course! With our supercharged team, we’re rapidly converging on a stellar Linkerd 2.17 release and fleshing out our long and very exciting roadmap beyond that. We’ll continue to add maintainers to the project and to grow our adopter base the only way we know how: by solving actual problems for actual people. (Without causing them more problems in the meantime.)

A handful of Linkerd maintainers and project participants will be in attendance at Kubecon NA in Salt Lake City this November. If you’re there, please do stop by the Linkerd booth in the project pavilion and say hi. All are welcome.

Needless to say, we’ll also continue feeling very grateful for our community. Thank you, again, from the bottoms of our hearts. We’d love to hear from you, whether about this change, your thoughts on the future of Linkerd, or any other mesh-y topics you’d like to discuss.

We honestly do want your input. Linkerd is unique in the service mesh space is that we are a single, nimble team working off a single, unified roadmap. This means that the voice of the adopter is one of the strongest forces in our universe.

Until our next update: onwards.

Linkerd is for everyone

Linkerd is a graduated project of 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!