What changed

Microsoft announced general availability on 24 August 2026. The announcement describes eBPF host routing as moving packet forwarding and routing decisions directly into the Linux kernel, removing the overhead of iptables-based processing and reducing dependence on the host networking stack, and states this results in lower latency, fewer network hops and higher throughput for application traffic. It also describes an enhanced enablement experience that respects existing network connections during rollout, performing a controlled node rotation to transition nodes to the new networking mode. The Learn documentation confirms and quantifies the shape of those benefits without publishing a percentage: reduced latency from bypassing iptables in the host, increased throughput for pod-to-pod traffic between nodes, and a modest reduction in CPU usage from removing iptables-based SNAT and routing logic. Microsoft names the intended use cases as performance-critical workloads such as high-throughput microservices, real-time services and AI or ML workloads, and adds the instruction to ensure the deployment environment meets the requirements before enabling. That sentence is the one worth taking literally, because the requirements are more restrictive than the announcement suggests.

The operational risk is easy to miss because the service can continue to look healthy. The control becomes visible only when a capacity request fails, an unsupported runtime is removed, or an extension blocks an enforced ERP update. Waiting for that moment transfers a planned decision into an incident.

What the term means in plain language

eBPF is a Linux kernel technology that runs small, verified programs inside the kernel itself. Host routing is the work a Kubernetes node does to decide where a packet goes, which by default is handled by iptables and netfilter rules in the host network namespace. eBPF host routing replaces that path with kernel-level eBPF programs, removing the iptables processing overhead. It is a feature of Advanced Container Networking Services, which is Microsoft’s chargeable suite of networking add-ons for AKS, and specifically of its Container Network Performance feature set. Azure CNI Powered by Cilium is the container networking mode that uses Cilium as the data plane, and it is a prerequisite.

This distinction matters because product status is not the same as business readiness. Availability, support and compatibility are separate questions. A service can be available but unsupported, supported but capacity-constrained, or technically updated while a customer-specific process has stopped working.

Why this matters to a South African organisation

South African teams often operate with tight specialist capacity, rand-sensitive budgets and business processes that cannot be paused while a replacement is sourced. Localisation, regional cloud capacity and long procurement lead times can narrow the recovery options. The practical response is to use the available test window before it becomes an emergency window.

The consequence belongs to the business process, not only the technology team. Finance month-end, customer transactions, data pipelines and ERP extensions all cross technical and operational ownership. A change should therefore be accepted only when the service owner and the business owner can see the same evidence.

The hidden exposure

eBPF host routing in Advanced Container Networking Services reached general availability on 24 August 2026, moving packet routing into the Linux kernel for lower latency and higher throughput. What the announcement does not say is that once it is enabled, AKS blocks anything from installing iptables rules in the host network namespace, that bypassing the block may leave the cluster inoperational, that it is all or nothing across the cluster, and that the service it belongs to is chargeable.

Normal operation is weak evidence. It proves only that yesterday’s combination of platform, configuration and workload completed. It does not prove that the next capacity allocation, lifecycle enforcement or major release will preserve the same result. An owner needs an inventory, a representative test and a dated decision.

Decision path

This is not a question of whether faster is better. It is a question of what else on the node depends on the path being removed. Microsoft states that enabling eBPF host routing causes iptables rules in the host network namespace to be bypassed, and that on clusters with it enabled, AKS blocks attempts to install iptables rules in the host network namespace. The consequence is stated in unusually direct language for a documentation page: trying to bypass this block may cause the cluster to be inoperational. Microsoft has built in a guard, stating that AKS attempts to detect and block enablement on clusters where iptables rules are already in use in the host network namespace, but attempts is the operative word and a guard is not a substitute for knowing what is running. The practical exposure is anything that writes host-namespace iptables rules, which in most estates means a security or endpoint agent, a network monitoring agent, or a custom DaemonSet written in-house. Braintree’s view, not Microsoft’s, is that the inventory is the whole job here: the enablement itself is a single operation, and the risk sits entirely in what nobody remembered was installed. Four further constraints narrow the field before that conversation even starts. It requires Azure CNI Powered by Cilium, because Container Network Performance is available only on those clusters. It is all or nothing: Microsoft states eBPF host routing can only be enabled for all nodes in a cluster and that hybrid node scenarios are not supported. Node operating systems are limited to Ubuntu 24.04 or Azure Linux 3.0, with Confidential VMs and Pod Sandboxing unsupported, and Windows nodes excluded because Azure CNI Powered by Cilium does not support them. And it cannot be used alongside Static Egress Gateway. There is also a cost that the announcement never mentions and the documentation states under its own heading: Advanced Container Networking Services is a paid offering. Against all of that sits one genuinely reassuring fact. It is reversible, and Microsoft has engineered for that: the IP masquerade agent remains running while eBPF host routing is active, with its iptables rules ignored, specifically to maintain consistent behaviour if the feature is later disabled.

Record the alternatives that were rejected and why. That prevents the next reviewer from reopening the entire question without context. Where the preferred path cannot be completed inside seven days, approve a time-bound exception with a responsible owner, expiry date and compensating control.

Technical test plan

The enablement itself performs a rolling upgrade rather than an in-place switch. Microsoft states that enabling eBPF host routing updates nodes by rolling upgrades for node pools, respecting existing connections through the node drain timeout, and that nodes are labelled kubernetes.azure.com/ebpf-host-routing=true once it is enabled on a given node. That label is the cheapest way to verify the rollout reached every node rather than assuming it did. Plan it as a maintenance operation with a change window, because a full node rotation on a large cluster is not a short event and the drain timeout governs how gracefully long-lived connections are handled. Two components change behaviour once it is active. An init container named iptables blocker prevents future installation of iptables rules in the host network namespace. The IP masquerade agent keeps running, but Cilium takes over SNAT using BPF-based masquerading and the agent’s iptables rules are ignored while the feature is active. Before enabling, audit the host network namespace on a representative node from each pool and record what writes rules there; treat any third-party agent as in scope until its vendor confirms otherwise in writing, because the failure mode is a cluster that stops working rather than a component that logs a warning. Confirm the prerequisites mechanically rather than from memory: CNI mode, node image and operating system version per node pool, whether Static Egress Gateway is configured, and whether any node pool uses Confidential VMs or Pod Sandboxing. Dual-stack connectivity is supported, so an IPv4 and IPv6 cluster is not itself a blocker. Finally, measure. The justification for enabling this is a performance improvement, and Microsoft publishes no percentage, so the only way to know whether it was worth the change and the ongoing charge is a before-and-after measurement of pod-to-pod latency and throughput on the same workload.

Use production-representative conditions without exposing production data unnecessarily. Capture the starting configuration, exact version, time of test and expected result. A pass requires evidence from the real workflow, not only a successful login or an unchanged dashboard.

Primary owner

Primary owner: The Kubernetes platform lead, with the owner of any security or monitoring agent running on the nodes.

The named owner coordinates platform, application, commercial and business-process decisions. Contributors may perform the work, but accountability cannot be distributed across a meeting invite. The owner closes the test, exception and evidence record.

Action within seven days

Action within seven days: Before enabling anything, list what installs iptables rules in the host network namespace on your AKS nodes, including security agents, monitoring agents and any custom DaemonSet. Confirm the node operating system is Ubuntu 24.04 or Azure Linux 3.0, confirm the cluster runs Azure CNI Powered by Cilium, and confirm Static Egress Gateway is not in use.

Start with the highest-consequence workload. Assign the people, date and pass criteria before the test begins. If the first test fails, record the failure as evidence and open remediation with a deadline rather than hiding it behind a general project status.

Evidence to retain

Evidence to retain: The dated inventory of host-namespace iptables consumers per node pool, the node operating system and CNI mode for each cluster, a written confirmation from each agent owner that their component does not depend on host iptables rules, and a before-and-after latency or throughput measurement taken on the same workload.

Store the evidence with the platform or change record. Include source exports and machine-readable results where possible. The next reviewer should be able to reproduce the conclusion without rebuilding it from email, chat or memory.

Frequently asked questions

What actually changes on the node?

Routing decisions move from iptables and netfilter processing in the host network namespace into eBPF programs in the kernel, and Cilium takes over SNAT using BPF-based masquerading.

What is the catch?

Once enabled, AKS blocks attempts to install iptables rules in the host network namespace, and Microsoft states that trying to bypass this block may cause the cluster to be inoperational.

Can we enable it on one node pool first?

No. Microsoft states it can only be enabled for all nodes in a cluster and that hybrid node scenarios are not supported.

What are the prerequisites?

Azure CNI Powered by Cilium, and node operating systems limited to Ubuntu 24.04 or Azure Linux 3.0. Confidential VMs and Pod Sandboxing are not supported, Windows nodes are excluded, and it cannot be used with Static Egress Gateway.

Does enabling it disrupt running workloads?

It performs a rolling node pool upgrade that respects existing connections through the node drain timeout. It is a maintenance operation rather than a configuration toggle.

Is it reversible?

Yes. Microsoft keeps the IP masquerade agent running while the feature is active, specifically to maintain consistent behaviour if eBPF host routing is later disabled.

Does it cost anything?

Yes. Microsoft states that Advanced Container Networking Services is a paid offering. The general availability announcement does not mention pricing.

How much faster is it?

Microsoft publishes no percentage, describing lower pod-to-pod latency, higher throughput between nodes and a modest CPU reduction. Measure your own workload before and after rather than assuming a figure.

The Braintree view

Microsoft’s announcement supplies the platform fact. The customer control begins after that fact: identify the exposed process, name the owner, test the real dependency and retain a decision that can survive audit or staff turnover. Braintree can help structure the inventory, build the representative test and translate the result into a controlled implementation plan.

Use the seven-day action as the entry point. Do not wait for a renewal, support refusal or enforced update to reveal work that can be measured now.