Note: this is an expanded version of the post on linkerd.io with additional details.

Today, we're happy to announce the release of Linkerd 2.17, a new version of Linkerd that introduces several major new features to the project: egress traffic visibility and control; rate limiting; and federated services, a powerful new multicluster primitive that combines services running in multiple clusters into a single logical service. This release also updates Linkerd to support OpenTelemetry for distributed tracing.

Linkerd 2.17 is our first major release since our announcement of Linkerd's sustainability and Buoyant's profitability in October. Not unrelatedly, it is one of the first Linkerd releases in years to introduce multiple significant features at once. Despite this, we worked hard to stay true to Linkerd's core design principle of simplicity. For example, these new features are designed to avoid configuration when possible; and when not possible, to make it minimal, consistent, and principled. After all, Linkerd's simplicity—our rejection of the status quo that says, "the service mesh is complex and must be complex"—is key to its popularity, and it's our duty to live up to that reputation in this and every release.

Read on for more!

Egress visibility and control

Linkerd 2.17 introduces visibility and control for egress traffic leaving the Kubernetes cluster from meshed pods. Kubernetes itself provides no mechanisms for understanding egress traffic, and only rudimentary ones for restricting it, limited to IP ranges and ports. With the 2.17 release, Linkerd now gives you full L7 (i.e. application-layer) visibility and control of all egress traffic: you can view the source, destination, and traffic levels of all traffic leaving your cluster, including the hostnames, and, with configuration, the full HTTP paths or gRPC methods. You also can deploy egress security policies that allow or disallow that traffic with that same level of granularity, allowing you to allowlist or blocklist egress by DNS domain rather than IP range and port.

Linkerd's egress functionality does not require changes from the application and only minimal configuration to get started. For more advanced usage, egress configuration is built on Gateway API resources, allowing you to configure egress visibility and policies with the same extensible and Kubernetes-native configuration primitives used for almost every other aspect of Linkerd, including dynamic traffic routing, zero trust authorization policies, and more. 

For example, enabling basic egress metrics across the entire cluster is as simple as adding this configuration:

apiVersion: policy.linkerd.io/v1alpha1
kind: EgressNetwork
metadata:
  namespace: linkerd-egress
  name: all-egress-traffic
spec:
  trafficPolicy: Allow

Rate limiting

Rate limiting is a reliability mechanism that protects services from being overloaded. In contrast to Linkerd's circuit breaking feature, which is client-side behavior designed to protect clients from failing services, rate limiting is server-side behavior: it is enforced by the service receiving the traffic and designed to protect it from misbehaving clients.

Just as with egress, Linkerd's rate limiting feature is designed to require minimal configuration, while still being flexible and configurable to a wide variety of scenarios. For example, a basic rate limit of 100 requests per second for a Server named "web-http" can be enabled with this configuration:

apiVersion: policy.linkerd.io/v1alpha1
kind: HTTPLocalRateLimitPolicy
metadata:
  namespace: emojivoto
  name: web-rlpolicy
spec:
  targetRef:
    group: policy.linkerd.io
    kind: Server
    name: web-http
  total:
    requestsPerSecond: 100

Linkerd's rate limiting feature also provides per-client rate limit policies that allow you to ensure rate limits are distributed "fairly" across multiple clients.  Combined with retries, timeouts, circuit breaking, latency-aware load balancing, and dynamic traffic routing, rate limiting extends Linkerd's already wide arsenal of in-cluster distributed system reliability features.

Federated services

In Linkerd 2.17 we've shipped an exciting new multicluster feature: federated services. A federated service is a logical union of the replicas of the same service across multiple clusters. Meshed clients talking to a federated service will automatically load balance across all endpoints in all clusters, taking full advantage of Linkerd's best-in-class latency-aware load balancing.

With federated services, not only is application code decoupled from cluster deployment decisions—service Foo talking to service Bar needs only to call "Bar", not to specify which cluster(s) it is on—but failure handling is transparent and automatic as well. Linkerd will transparently handle a wide variety of situations, including:

• An entire cluster is down
• A cluster is up the service on that cluster is down, or failing in some way (including L7 failures, e.g. returning 5xx response codes)
• A cluster is slow, or the service is slow

In all these cases, Linkerd will automatically load balance across all service endpoints on all clusters, using its default latency-aware (latency EWMA) balancing to send individual requests to the best endpoint.

Federated services were designed to capture a recent trend we see in multicluster Kubernetes adoption: planned large-scale multicluster Kubernetes. Linkerd's original multicluster functionality, released in the good ol' days of Linkerd 2.8, was designed for the ad-hoc, pair-to-pair connectivity that was common at the time. However, modern Kubernetes platforms are often much more intentional in their multicluster usage, sometimes ranging into the hundreds or thousands of clusters. Federated services join features such as flat network / pod-to-pod multicluster (introduced in Linkerd 2.14) in the toolbox for this new class of Kubernetes adoption.

Linkerd Day at Kubecon London

Next April we're happy to report that the CNCF is hosting Linkerd Day at Kubecon London. Many of the Linkerd maintainers will be in attendance, and we're expecting a great lineup of Linkerd talks as well as plenty of Linkerd users. Come see us in London!

Getting started with Linkerd 2.17 today!

The latest stable release package of Buoyant Enterprise for Linkerd, BEL 2.17.0, is now available, with a rich Linkerd 2.17 changelog. BEL is free for anyone to download and use in non-production environments, and free for companies with fewer than 50 employees to run in production.  BEL is the distribution of Linkerd that we run ourselves. Get started with BEL in under five minutes!

We're also doing a hands-on workshop on Linkerd 2.17 next week Thursday, December 12. Register today!