AKS Fleet Manager Placement Controls and Upgrade Health Gates: Enough Multi-Cluster Governance for a Small Platform Team?
At KubeCon North America 2026, Microsoft made Azure Kubernetes Fleet Manager placement controls generally available and announced upgrade health gates and upgrade insights. Here is what Fleet Manager now governs across AKS clusters, what it leaves out (app delivery, environments, previews, per-app access, agent governance), and when a small platform team can stop there.
The announcement is real and mostly about clusters. On November 10, 2026, Microsoft made three Azure Kubernetes Fleet Manager placement controls generally available (drift detection, takeover and eviction with a disruption budget), put annotation-based placement in public preview, and announced upgrade insights and upgrade health gates, which stop a rollout when node, cluster or workload health deteriorates.
Fleet Manager governs cluster upgrades and Kubernetes resource placement well. Update runs with stages, approvals, soak times and auto-upgrade profiles cover up to 1,000 member clusters, and the Fleet Manager resource itself is free; you pay for the hub cluster.
It does not govern application delivery. Fleet Manager does not build images from Git, model environments (app plus database plus config), create preview environments per pull request, decide who may deploy which service to production, or scope what an AI agent may do.
Fleet Manager alone is enough for a small team running a handful of AKS clusters with one or two apps, a working GitOps or CI pipeline, and no per-team or per-PR environments. Past that, you add a deployment layer: Flux or Argo CD you assemble, or a platform such as Qovery.
Watch the edges. Multi-cluster updates fully support AKS only (AKS Automatic and node auto-provisioning are partial, Arc-enabled clusters are unsupported), and AKS platform-initiated upgrades bypass Fleet Manager update runs entirely.
KubeCon North America gave Azure users a good week. Fleet Manager picked up GA placement controls, health-aware upgrade gates and pre-upgrade checks, and the question I keep hearing from Azure teams since is simple: "Is this enough multi-cluster governance for us, or do we still need something on top?"
Many of the platform teams I work with on Azure are three to eight people. For them, the answer depends on what "governance" means in their organization. If it means "our clusters upgrade in a safe order and the same baseline config lands everywhere," Fleet Manager now covers most of it. If it means "we control who ships what, where, and we can prove it," Fleet Manager covers very little, because it was never built to. Here is the line between the two, with the sources.
What did Microsoft announce for AKS Fleet Manager at KubeCon NA 2026?
Microsoft made three Fleet Manager placement controls generally available (drift detection, takeover and eviction), released annotation-based placement in public preview, and announced upgrade insights and upgrade health gates for safer automated upgrades across clusters. The details come from Brendan Burns' KubeCon post of November 10, 2026.
Here is each piece, quoted from Microsoft:
Drift detection (GA) "flags resources that no longer match their placement configuration."
Takeover (GA) "lets Fleet Manager adopt workloads on a joining cluster under rules the platform team sets."
Eviction (GA) "removes resources from a member cluster without connecting to it, and a disruption budget caps how many placements go offline at once."
Annotation-based placement (public preview) "declares target clusters with annotations on a standard Kubernetes resource instead of a separate placement object."
Upgrade insights "checks readiness before a cluster or node image upgrade begins and flags issues like restrictive pod disruption budgets, deprecated APIs, and quota or subnet limits."
Upgrade health gates evaluate "node, cluster, and workload health, including signals from your own monitoring tools, and stops the upgrade from progressing when health deteriorates."
Check the status before you write any of this into a runbook. The AKS engineering blog describes health-aware upgrade gates as preview and advises you to "check current support and requirements before making either part of your contract," and the Pixel Robots recap of Azure Day notes that everything on the upgrade-fixes slide was marked preview. Placement controls are the GA part.
Fleet Manager is built on KubeFleet, a CNCF Sandbox project accepted on January 21, 2025. Burns summed up the direction with one line: "Kubernetes is at its best when the system, rather than a person, keeps what's running matched to what you asked for" (Cloud Native Now).
What does Azure Kubernetes Fleet Manager cover today?
Fleet Manager covers two jobs: orchestrating Kubernetes and node image upgrades across many clusters, and placing Kubernetes resources from a hub cluster onto member clusters. Around those, it adds multi-cluster namespaces with quotas and access, cross-cluster networking, and several preview features.
Membership. AKS clusters across regions and subscriptions, plus Arc-enabled Kubernetes clusters "across clouds and on-premises." Up to 1,000 member clusters per fleet, all in the same Microsoft Entra ID tenant.
Update orchestration. Update runs apply Kubernetes and node image upgrades in stages and groups, with reusable strategies, manual or automated approvals, after-stage waits and scheduled start gates. Auto-upgrade profiles trigger runs when new versions publish.
Resource placement (needs a hub).ClusterResourcePlacement propagates cluster-scoped resources and whole namespaces, and ResourcePlacement propagates specific namespace-scoped objects such as ConfigMaps, Secrets or Deployments. The AKS blog points out you can roll those resources out "stage by stage, with soak times and approval tasks between stages."
Managed Fleet Namespaces. One namespace defined centrally, placed on chosen clusters, with CPU and memory quotas, a built-in network policy and four built-in Azure RBAC roles (docs).
Previews. Automated Deployments that stage resources from Git to the hub, DNS-based load balancing across AKS clusters, and identity binding profiles for workload identity.
Cost is easy to reason about. The Fleet Manager pricing page says the resource "is free to use and you only pay for the single node uptime SLA Azure Kubernetes Service cluster and its infrastructure resources like virtual machine (Standard_DS3_v2 SKU) and load balancers." Without a hub, you get update orchestration and pay nothing extra. With a hub, you pay for one small AKS cluster that Microsoft manages and keeps upgraded for you.
Where does Fleet Manager stop?
Fleet Manager stops at the Kubernetes resource. It moves manifests and upgrades clusters, and it has no concept of an application, a build, an environment, a pull request or an agent. It also has hard edges on which clusters it can upgrade.
Upgrade coverage has gaps you should know before relying on it
The Fleet Manager FAQ has a support table for multi-cluster updates:
Cluster type
Multi-cluster updates
Detail from Microsoft
AKS in Azure
Supported
Full support
AKS Automatic
Partial
"You can't disable cluster-level auto-upgrade, so the cluster can update out of sequence"
AKS with node auto-provisioning
Partial
"Only Kubernetes control plane upgrades are supported"
AKS on bare metal, Edge Essentials, Azure Local
Unsupported
On the roadmap
Arc-enabled clusters (EKS, GKE, on-premises)
Unsupported
On the roadmap
So an Arc-connected EKS or GKE cluster can join the fleet and receive placed resources, but Fleet Manager will not upgrade it. Your multi-cloud upgrade story stays per-cloud.
Two more limits matter for small teams:
Platform-initiated upgrades bypass Fleet. If a member cluster stays on an unsupported version past its grace period, AKS upgrades it directly, and "Fleet Manager update groups and strategies don't control the upgrade sequence, and the upgrade can happen outside the cluster's maintenance window" (FAQ). AKS starts these on June 1, 2027. Our AKS LTS and Kubernetes 1.37 upgrade plan walks through the calendar.
Fleet Manager cannot create clusters. "Creation and lifecycle management of new AKS clusters is on our roadmap." You still provision clusters with Terraform, Bicep or the portal. Fleet Manager is also a regional resource; region failover for it is on the roadmap.
Application delivery is outside its scope
Fleet Manager places Kubernetes objects that already exist. Something else has to turn a Git commit into an image, write the Deployment, wire the database connection string, and decide which environment gets it. Microsoft's own guidance is to "separate infrastructure upgrades from Kubernetes resource rollouts" and to pair Fleet with Flux for GitOps (AKS blog). That pairing works, and you own both halves of it.
Access control stops at the namespace
Managed Fleet Namespaces give you four roles: Reader, Writer, Admin and Cluster Admin for member clusters. That is real multi-cluster access control, and it is namespace-shaped. A Writer can read Secrets "and can assume any ServiceAccount in the namespace," role assignments at a namespace scope also grant access to unmanaged namespaces with the same name on member clusters, and "access controls for Kubernetes Custom Resources (CRs) are not supported for these roles" (docs). There is no way to say "Alice may deploy the payments service to staging, but only approve production."
Which governance needs does Fleet Manager meet, and which need a deployment layer?
Fleet Manager meets cluster-level needs: safe upgrade order, consistent baseline resources, drift detection, namespace quotas and namespace access. Application-level needs (who deployed which commit where, preview environments, environment cloning, per-service access and agent limits) need a deployment layer on top.
Governance need
Fleet Manager
What you add
Typical choice
Upgrade clusters in a safe order with approvals
Yes, update runs and strategies
Nothing on AKS
Fleet Manager
Stop an upgrade when health drops
Announced (health gates, check preview status)
Health signals from your monitoring
Fleet Manager plus Azure Monitor or Prometheus
Same baseline (policies, namespaces, config) on every cluster
Yes, with a hub
Nothing
Fleet Manager placement
Detect and correct config drift
Yes, drift detection (GA)
Nothing
Fleet Manager
Upgrade non-AKS clusters
No (Arc unsupported for updates)
Per-cloud upgrade tooling
EKS or GKE native, or a managed platform
Build images from Git and deploy on merge
No
CI plus CD
GitHub Actions plus Flux or Argo CD, or Qovery
Environments as a unit (app, database, variables)
No
Environment model
Helm plus Terraform, or Qovery
Preview environment per pull request
No
Ephemeral environments
Argo CD PR generator, or Qovery
Per-team, per-service, per-environment deploy rights
Namespace roles only
App-level RBAC
Platform RBAC (Qovery, Argo CD projects)
Audit trail of who deployed which commit where
Azure activity logs for Fleet actions
Deployment history per service
Git history plus CD logs, or Qovery audit logs
Limit what an AI agent may deploy
No
Scoped agent identity and policy
OPA-based policy tokens, or per-agent RBAC
Cost per environment and team
No
Cost allocation
Azure Cost Management tags, or Qovery cost visibility
The top rows are cluster operations, where Fleet Manager has become a strong default on Azure. The bottom rows are the questions a developer or an auditor asks, and Fleet Manager does not answer them.
Fleet Manager upgrades the clusters. Who ships the apps?
Qovery adds git-push deployments, preview environments, per-environment RBAC and audit logs on AKS in your own Azure subscription, or on clusters you already run. Start on the Free plan.
When is Fleet Manager alone enough for a small platform team?
Fleet Manager alone is enough when your clusters are all AKS, you deploy one or two applications that change through an existing CI or GitOps pipeline, and nobody asks for per-team or per-pull-request environments. In that setup, the hard problem is keeping clusters current and consistent, and Fleet Manager solves it for free.
Concretely, I tell teams they can stop at Fleet Manager if most of these are true:
All member clusters are AKS in Azure, not AKS Automatic with forced auto-upgrade, not Arc-connected EKS or GKE.
Two to ten clusters, split by environment or region (for example dev, staging, prod-EU, prod-US).
One or two services, deployed by a pipeline that already works and that the team understands.
Access needs are namespace-shaped: team A owns namespace A everywhere, and that is the whole policy.
No preview environments per pull request, and no request for them on the roadmap.
No AI agents with write access to production infrastructure.
In that case, set up update runs with a dev-then-prod strategy and approvals, put a hub in when you need the same namespaces and policies everywhere, enable drift detection, and keep Flux reconciling application manifests on each cluster. You will spend your time on the cluster layer, which is where a small team on Azure should spend it. Our comparison of AKS and EKS for a small DevOps team lists the other add-ons you still run on either cloud (ingress, certificates, autoscaling, secrets, observability).
When do you still need a deployment layer on top of Fleet Manager?
You need a deployment layer on top of Fleet Manager as soon as the governance question moves from "are our clusters safe?" to "who can ship what, where, and can we prove it?" That usually happens with the third product team, the first audit, the first request for preview environments, or the first AI agent with deploy rights.
The signals I watch for:
Developers wait on the platform team to create namespaces, environments or pipelines.
You want a full-stack environment per pull request, with its own database and URL, destroyed on merge.
An auditor asks for a list of every production deployment last quarter with the commit, the person and the approval.
Teams need different rights per environment: deploy to staging freely, production behind approval.
Coding agents (Claude Code, Cursor, Codex) are starting to deploy, and you need them scoped and attributed.
Some clusters are not AKS (an EKS cluster from an acquisition, an on-premises cluster), and you want one way to deploy to all of them.
You then have two realistic paths:
Path
What you assemble
Strength
Cost for a small team
Fleet Manager plus Flux or Argo CD plus CI
Fleet for clusters, GitOps for manifests, GitHub Actions or Azure Pipelines for builds, Argo CD ApplicationSets for previews, Kyverno or Gatekeeper for policy
Fully open source, Azure-native for clusters
You own and upgrade five or more tools, and the glue between them
Fleet Manager plus a deployment platform (for example Qovery)
Fleet for AKS upgrades if you run your own clusters, the platform for apps, environments, previews, RBAC and audit
What does a sensible setup look like for a four-person platform team on AKS?
A sensible setup uses Fleet Manager for upgrade order and baseline resources, a deployment layer for applications, and one rule that keeps them from fighting: Fleet Manager never places application workloads, and the deployment layer never touches cluster versions or shared baseline objects.
A layout I would recommend:
Clusters. Provision AKS with Terraform. Standard tier is $0.10 per cluster-hour; Premium, needed for LTS, is $0.60 (AKS pricing).
Fleet Manager without a hub first. Join every cluster, disable per-cluster auto-upgrade so Fleet owns the order (the FAQ warns that otherwise "either Fleet Manager or AKS cluster auto-upgrade performs the update, depending on which one runs first"), and create a strategy: dev, then staging after a soak, then production behind an approval.
Add upgrade insights and health gates once their status fits your risk appetite, wired to the monitoring you already trust.
Add a hub only when you need placement, for example the same network policies, quotas and namespaces on every cluster. Turn on drift detection.
Applications through one deployment layer, with environments, previews and RBAC defined there, not in Fleet placement objects.
Never let a cluster fall out of support, or platform-initiated upgrades will ignore your carefully designed strategy.
Where does Qovery fit next to Azure Kubernetes Fleet Manager?
Qovery is an Agentic DevOps platform that covers the application half of that table on AKS and any other cluster you run, while Fleet Manager keeps covering the cluster half. Your apps run on Kubernetes in your own Azure subscription (or AWS, GCP or Scaleway account), and Qovery's control plane stays separate from that data plane, so Qovery never hosts them.
There are two ways to run it on Azure:
Qovery-managed AKS in your subscription. Qovery provisions and upgrades the cluster for you, so you may not need Fleet Manager update runs for those clusters at all (announcement).
Your existing AKS clusters. Clusters you already run, including those in a Fleet, connect with qovery cluster install (BYOK docs). Fleet Manager keeps orchestrating their upgrades; Qovery handles what gets deployed on them.
On top, your developers get git-push deployments, environments that bundle services, databases and variables, environment cloning, a preview environment per pull request, RBAC per project and environment, and audit logs of every deployment and configuration change. For AI agents, Qovery's OPA-based API policy tokens limit exactly what an agent may do, and every agent action is attributed in the audit log.
Qovery does not replace Fleet Manager's multi-cluster resource placement, drift detection for arbitrary Kubernetes objects, or cross-cluster networking. If those are your whole problem, Fleet Manager is the better tool. The Free plan (3 users, unlimited AI agents, one cluster of your own, 1,000 deployment minutes a month) is enough to connect one AKS cluster and check whether the application layer fits.
How do I set this up with Qovery from Claude Code or Cursor?
Install the Qovery Agent Skills and, if you want your agent to read and operate existing environments, the Qovery MCP Server. Then ask the agent in plain language. These are the commands:
BASH|1. Agent Skills for every compatible agent (Claude Code, Cursor, Codex, OpenCode, Gemini CLI)
curl -fsSL https://skill.qovery.com/install.sh | bash
# 2. Or install the skills for the current project only
curl -fsSL https://skill.qovery.com/install.sh | bash -s -- --project
# 3. MCP Server in Claude Code over OAuth, read-only by default
claude mcp add --transport http qovery https://mcp.qovery.com/mcp --callback-port 4242
# 4. Write access, only when you explicitly want it (also enable write access in the Qovery Console)
claude mcp add --transport http qovery "https://mcp.qovery.com/mcp?read_write=true" --callback-port 4242
Step 1 asks you to approve a shell install. Step 3 opens an OAuth login, and the MCP Server then acts with your own Qovery permissions, read-only unless you choose step 4 and enable write access in the Console.
Then ask: "deploy this app to my Azure subscription with Qovery, with a staging and a production environment" or "set up preview environments for pull requests on this repo." The skills (qovery-onboard, qovery-deploy, qovery-preview and others) do the setup; the MCP Server operates what exists. Start from the AI agent quickstart or the Claude Code integration. Our Kubernetes operations hub covers upgrades, autoscaling and multi-cluster work across AKS, EKS and GKE.
Frequently asked questions
What did Azure announce for Kubernetes Fleet Manager at KubeCon NA 2026?
On November 10, 2026, Microsoft made three Fleet Manager placement controls generally available: drift detection, takeover of workloads on a joining cluster, and eviction with a disruption budget. It also released annotation-based placement in public preview and announced upgrade insights, which check readiness before an upgrade, and upgrade health gates, which stop an upgrade when node, cluster or workload health deteriorates. Microsoft's AKS blog describes health-aware upgrade gates as preview, so check their status before depending on them.
Are AKS Fleet Placement Controls and Upgrade Health Gates enough multi-cluster governance for a small platform team?
They are enough if your governance problem is cluster operations: safe upgrade order, consistent baseline resources and drift detection across AKS clusters. They are not enough if you need to control and audit application delivery, such as who deployed which commit to production, preview environments per pull request, per-service access rights or limits on AI agents. For those, you add a deployment layer such as Flux or Argo CD with CI, or a platform like Qovery.
Can Fleet Manager upgrade EKS or GKE clusters?
No. Arc-enabled clusters, including EKS and GKE connected through Azure Arc, can join a fleet as members and receive placed resources, but Fleet Manager multi-cluster updates do not support them. Microsoft lists Arc and AKS on bare metal or Azure Local as unsupported for updates, with support on the roadmap.
How much does Azure Kubernetes Fleet Manager cost?
The Fleet Manager resource is free. If you add a hub cluster for resource placement, you pay for that single-node AKS cluster with the uptime SLA and its infrastructure, such as a Standard_DS3_v2 virtual machine and load balancers. Microsoft manages and upgrades the hub cluster for you.
Does Fleet Manager replace Argo CD or Flux?
Not for application delivery. Fleet Manager places Kubernetes resources from a hub onto member clusters and can stage those rollouts with approvals, and Microsoft's own guidance pairs it with Flux for GitOps on each cluster. Fleet Manager does not build images from Git, model environments or create preview environments per pull request.
Do AKS platform-initiated upgrades respect Fleet Manager update strategies?
No. If a member cluster stays on an unsupported Kubernetes version past its grace period, AKS upgrades it directly, outside Fleet Manager update runs and possibly outside the maintenance window. Platform-initiated upgrades start on June 1, 2027, so keeping every member cluster on a supported version through Fleet update runs is the only way to keep control of the order.
Guillaume is a Staff Solutions Architect at Qovery. He helps customers troubleshoot production infrastructure and writes about Kubernetes operations, AI agents, and reliable cloud platforms.
Next step
Fleet Manager upgrades the clusters. Who ships the apps?
Qovery adds git-push deployments, preview environments, per-environment RBAC and audit logs on AKS in your own Azure subscription, or on clusters you already run. Start on the Free plan.