Prepare and test your skills

Prepare and test your skills

Network isolation in Google Kubernetes Engine (GKE) controls how Pods, nodes, and the control plane communicate so legitimate traffic is allowed and unwanted traffic is blocked. Google Kubernetes Engine (GKE) uses the VPC network, firewall rules, and visibility settings to enforce that isolation consistently across the cluster.
A GKE cluster consists of at least one control plane machine and multiple worker machines called nodes. The nodes are Compute Engine virtual machine (VM) instances that run the Kubernetes processes needed to join the cluster. You deploy applications to the cluster, and those applications run on the nodes while the control plane manages them.
intranode visibility (opens in a new tab) is a GKE networking setting that makes Pod-to-Pod traffic visible to the VPC network, even for Pods on the same node. When intranode visibility is enabled, the VPC network processes all packets sent between Pods, including packets that never leave a single node. This allows VPC firewall rules and hierarchical firewall policies to apply consistently to Pod-to-Pod communication regardless of Pod location. If you use ip-masq-agent with intranode visibility, your custom nonMasqueradeCIDRs list must include the cluster's node and Pod IP address ranges to prevent connectivity issues.
VPC firewall rules are the enforcement point for isolated Pod communication when intranode visibility is enabled. Because all Pod-to-Pod packets flow through the VPC network, custom ingress and egress allow rules must be evaluated carefully against the cluster's networking needs. You can use connectivity tests to verify that legitimate traffic is not obstructed by a new rule set. Pod-to-Pod communication must remain allowed for network policy to function correctly.
Enabling intranode visibility on a cluster activates VPC-level enforcement for traffic inside the cluster. For a new cluster, create it with the --enable-intra-node-visibility flag as gcloud container clusters create CLUSTER_NAME --location=CONTROL_PLANE_LOCATION --enable-intra-node-visibility. For an existing cluster, update it with the same flag, which restarts components in both the control plane and the worker nodes. You can also enable the setting in the Google Cloud console with this ordered workflow:
Exam tip: Enabling intranode visibility on an existing cluster recreates the nodes, which can disrupt running workloads.
GKE Standard clusters can be created as zonal clusters or regional clusters, and the choice controls where the control plane and worker nodes run. A zonal cluster keeps its control plane in one zone, while a regional cluster replicates its control plane across multiple zones in a region. GKE Autopilot clusters are always regional, so this choice applies to GKE Standard mode.
Side-by-side system architecture comparing a zonal GKE cluster with one control plane in one zone versus a regional cluster with replicated control planes and nodes across three zones behind a load balancer.
A zonal cluster (opens in a new tab) is a GKE cluster with a single control plane instance running in a single Compute Engine zone. The nodes for a zonal cluster can be placed in one zone, called a single-zone cluster, or spread across multiple zones in the same region, called a multi-zonal cluster. In a multi-zonal zonal cluster, all node zones must be in the same region as the control plane, and the list of node locations must include the control plane's primary zone.
Because the control plane exists in only one zone, a zonal cluster has a single point of failure for cluster management. If that zone becomes unavailable, the Kubernetes API becomes unavailable, even if multi-zonal nodes in other zones keep running workloads. Multi-zonal nodes improve workload availability compared to single-zone nodes, but they do not protect the control plane. After a cluster is created, it cannot be changed from zonal to regional or from regional to zonal.
A regional cluster (opens in a new tab) is a GKE cluster created in a region that replicates its control plane on virtual machines (VMs) situated in multiple zones within that region. Nodes communicate with the control plane through a load balancer, so a node's zone and the control plane VM that serves it do not always match. By default, a regional cluster is also multi-zonal for nodes, which means GKE distributes worker nodes evenly across multiple zones in the same region.
For example, a regional cluster in us-east1 creates control plane replicas in different us-east1 zones and provisions nodes in us-east1-b, us-east1-c, and us-east1-d. The default node pool for a regional Standard cluster has nine nodes, with three nodes in each of three zones, and it can be reduced to one node per zone if needed. A regional cluster can also use a single-zone node pool, where the control plane is still replicated across zones but all nodes run in one zone and are not replicated elsewhere.
Control plane redundancy and node distribution together determine how a cluster handles a zone failure. In a regional cluster, loss of one zone does not take down the control plane because redundant replicas in the surviving zones keep the Kubernetes API available during outages, upgrades, and resizes. In a zonal cluster, loss of the control plane zone interrupts API access, upgrades, and management operations until the zone recovers.
Workload availability depends on having node capacity and Pod replicas in the surviving zones. Regional multi-zonal placement keeps nodes in multiple zones, so workloads with sufficient replicas can continue running when one zone fails. Zone-simulation methods such as reducing node pool zones or cordoning and draining a zone's nodes affect only nodes, not the control plane. In Autopilot, GKE automatically rebalances nodes after an outage, while in Standard clusters nodes must be rebalanced manually or with the cluster autoscaler.
The two architectures differ in placement and failure behavior:
| Architecture | Control plane placement | Node placement |
|---|---|---|
| Zonal cluster | Single instance in one zone; zone failure interrupts the Kubernetes API | Single zone, or multiple zones in the same region; multi-zonal nodes help workloads but do not protect the control plane |
| Regional cluster | Replicated across multiple zones in one region; zone failure does not impact the control plane | By default spread evenly across three zones in the region; can optionally use a single-zone node pool with replicated control plane only |
Regional availability has tradeoffs in quota and networking cost. A regional Standard cluster uses more regional quota and IP addresses than a similar zonal cluster, and new billing accounts with only eight in-use IP addresses per region may need a quota increase to create the default nine-node regional cluster. Workloads that communicate across zones also incur cross-zone network charges when traffic flows from a node in one zone to a node in a different zone.
Exam tip: Choose a regional cluster for production workloads that need control plane redundancy, and remember that multi-zonal nodes alone do not make a zonal control plane highly available.
Google Kubernetes Engine (GKE) is a managed service for running containerized applications on Kubernetes clusters. A GKE cluster can be deployed in Autopilot mode and Standard mode, and the choice of mode affects how you plan compute resources and how much cluster configuration you handle directly.
A GKE cluster includes at least one cluster control plane machine and multiple worker machines called nodes. The nodes are Compute Engine virtual machine (VM) instances that run the Kubernetes processes necessary to make them part of the cluster. You deploy applications to clusters, and the applications run on the nodes. This same basic architecture underlies both Autopilot and Standard configurations.
GKE Autopilot mode is a cluster configuration that is created with a simplified command that specifies the cluster name and location. You create an Autopilot cluster with gcloud container clusters create-auto, for example gcloud container clusters create-auto $CLUSTER_NAME --region $LOCATION. Creation can take several minutes to complete, after which the cluster with its control plane and nodes is ready for workloads. The example events-cluster shows this pattern, where providing a name and region produces a complete regional GKE cluster.
Evaluating Autopilot versus Standard means weighing management responsibility, resource provisioning, and cost optimization for a specific workload type. With Autopilot, resource provisioning starts from the cluster name and region, and GKE assembles the control plane and node VMs that host your applications. With Standard, planning involves more direct decisions about the node pool, because the nodes that run your applications are Compute Engine VMs you configure and size. Workloads that benefit from minimal node-level planning align with the Autopilot creation model, while workloads with specific node requirements need closer planning of node capacity, availability, and underlying VM cost.
Exam tip: Use gcloud container clusters create-auto with a cluster name and --region to create an Autopilot cluster.
gcloud container clusters create-auto using a cluster name and region.A zonal cluster runs a single control plane instance in one zone, so loss of that zone makes the Kubernetes API unavailable, while a regional cluster replicates its control plane across multiple zones in one region so the API remains available during a zone failure. By default a regional cluster also spreads worker nodes evenly across three zones in the same region, whereas a zonal cluster places nodes in a single zone or across multiple zones in the same region without protecting the control plane.
Google Kubernetes Engine (GKE) Autopilot mode uses a simplified creation model that starts from the cluster name and region and always produces a regional cluster, which suits workloads that benefit from minimal node-level planning. Google Kubernetes Engine (GKE) Standard mode requires more direct decisions about the node pool because nodes are Compute Engine virtual machine (VM) instances that you configure and size, which suits workloads with specific node requirements for capacity, availability and underlying virtual machine cost.
Intranode visibility is a Google Kubernetes Engine (GKE) networking setting that makes Pod-to-Pod traffic visible to the VPC network even for Pods on the same node, so VPC firewall rules and hierarchical firewall policies apply consistently regardless of Pod location. Enabling it on an existing cluster restarts components in both the control plane and the worker nodes and recreates the nodes, which can disrupt running workloads.
An organization is deploying general-purpose containerized microservices to Google Kubernetes Engine (GKE). The platform engineering team has limited operational capacity and wants to eliminate the administrative overhead of managing underlying node pools, planning virtual machine capacity, and manually configuring Pod bin-packing.
They require an operational model where GKE automatically provisions and manages compute infrastructure according to Pod resource requests while enforcing security best practices by default.
Which GKE mode of operation should the organization deploy?
Last updated