Skip to content

Google makes sidecar-less ambient networking generally available on GKE

Google Cloud has made ambient networking for Google Kubernetes Engine generally available, offering service mesh features without per-pod sidecar proxies. The company claims up to a 10x reduction in resource use for L4 mesh features.

3 min readBased on 1 source0 views
CLOUD

TL;DR

  • Google Cloud has made ambient networking for GKE generally available, a service mesh design that runs without sidecar proxies in each pod.
  • Shared per-node Envoy proxies handle L4 traffic, handling mTLS, authorization policies, load balancing and telemetry.
  • Google says the approach can cut resource usage up to 10x for L4 mesh features and reduce disruption from CVE patches.

What changed

Service meshes usually work by injecting a sidecar container into every application pod. That proxy handles encryption, access rules and metrics, but it also adds CPU and memory cost and extra work for platform teams. Google Cloud's announcement says its ambient networking for Google Kubernetes Engine (GKE) is now generally available, moving those duties out of the pods entirely.

In the ambient design, networking responsibilities shift from individual pods to shared components on each node and to managed control plane services. A node-level Envoy instance, which Google describes as an L4 node proxy, processes intercepted layer 4 traffic. Envoy is an open-source proxy widely used in service meshes. The node proxy enforces mutual TLS (mTLS), where both sides of a connection verify each other's certificates, evaluates authorization policies, performs load balancing and emits telemetry.

How the control plane fits in

The control plane is split across several pieces. Traffic Director acts as the centralized xDS control plane, pushing routing rules, access policies and cluster state to the node proxies. xDS is the family of discovery APIs that Envoy uses to receive configuration. Google's Certificate Authority Service issues managed workload identity certificates on demand. A Gateway controller reads Gateway API resources and passes the configuration to Traffic Director.

On each node, an NRI plugin uses the Kubernetes node resource interface to redirect traffic to and from ambient-enabled pods, and an xDS agent inside the node proxy pod receives configuration and requests certificates. Google says these components run in a dedicated gke-managed-ambient namespace.

Operations and upgrades

The most practical change for operators is lifecycle management. Because proxy logic no longer lives in application containers, upgrading proxy versions or pushing configuration changes does not require restarting workloads. Google says this avoids restarts and downtime windows, and that it can reduce administrative overhead and application disruption by an average of four times a year by removing pod restarts needed for L4 mesh CVE patches. The product can be installed at cluster creation or added to an existing cluster, and maintenance runs alongside normal GKE cluster maintenance.

Observability and security features

Ambient networking supports Zero Trust-style setups in which service-to-service traffic is encrypted with mTLS using managed identities. Authorization policies let operators define access rules that enforce least-privilege boundaries between microservices. Telemetry includes L4 metrics out of the box, and L7 metrics are also available. To view this data, Google points to Network Services Monitoring, a new Google Cloud capability released at the same time. It shows service-to-service connections, latency and error rates across GKE clusters without custom dashboards.

Why it matters

For teams running service meshes on GKE, the main trade-off has been the cost of sidecars in every pod. Ambient networking offers a way to keep mTLS, access policies and telemetry while removing that per-pod overhead. Google's headline figure of up to a 10x reduction in resource usage applies to L4 mesh features, and the company does not give a comparable figure for L7 processing in this announcement, so teams should measure their own workloads before planning migrations. Platform teams should also weigh the move to shared node proxies and managed control plane services, and check the GKE documentation and pricing page that Google links to.

Sources

This article was written by Blozhik from the reporting below. Read the originals for full detail.

GoIntroducing ambient networking for GKEGoogle Cloud Blog
Was this useful?

Keep reading

Discussion · 0

Sign in to join the discussion. Liking does not need an account.Sign inRegister
Loading comments…
blozhikDeveloper news from across the web, explained. Every story links to its original sources.