Prepare and test your skills

Prepare and test your skills

Hybrid and multi-cloud strategies describe where workloads run. A hybrid architecture combines on-premises infrastructure with public cloud, while a multi-cloud architecture uses services from more than one cloud service provider (CSP). Organizations choose these models for flexibility, compliance, and resilience. The tradeoff is added complexity and operating cost compared with running in one cloud.
Avoiding vendor lock-in means not depending on one provider for every workload. A multi-cloud approach lets a business pick the best service for each job, but feasibility depends on application dependencies, interoperability, rebuilding cost, technical skills, and consistent security across providers. Meeting data sovereignty means storing and processing sensitive data such as personally identifiable information (PII) in a required country or region. The common case is expansion into a country where the current CSP has no local region, so the organization uses another CSP that does have a local region there.
Strengthening reliability means keeping critical applications available if one provider fails. In a multi-cloud design, traffic can be routed from a failed CSP region to another CSP in the same region. This failover works only when both providers offer the required services in that region. Resilience planning must check four practical items:
Exam tip: Using two CSPs in the same region can meet data residency rules and improve availability at the same time, but only when both providers offer the needed services in that region.
Google Kubernetes Engine (GKE) (opens in a new tab) provides a common Kubernetes runtime for modernizing legacy applications across on-premises, edge, and cloud. Applications are packaged into containers and deployed as Pods to nodes, where nodes are Compute Engine virtual machines grouped into a cluster. Teams use the Kubernetes API to administer, scale, and monitor workloads, while Google Cloud manages the control plane and system components, and in Autopilot mode Google Cloud also manages worker nodes.
Kubernetes abstracts differences among environments so teams can develop, run, and move workloads consistently between edge and cloud. Portability still has limits because performance can differ due to underlying compute, networking, security, proximity to dependent services, and load balancer behavior. Moving a workload may also require moving its data, and for hybrid container setups Google Distributed Cloud connected can simplify setup and operation.
Consistent security in GKE Enterprise means the same controls apply to workloads, images, data, identities, and network traffic no matter where they run. Key controls include Policy Controller for policy enforcement, Config Sync for managing cluster configuration from a source of truth, the security posture dashboard, Container-Optimized OS, and automatic upgrades. Binary Authorization allows or blocks image deployment, OS scanning checks containers for vulnerabilities before deployment, and VPC Service Controls defines perimeters across GKE and Cloud Run resources. Data is protected with Sensitive Data Protection for discovery and classification, encryption in transit with VPN tunnels or TLS, and common identity between environments.
Service-to-service communication across hybrid environments uses controlled, authenticated, and encrypted paths. GKE supports a zero trust distributed architecture where services on-premises or in another cloud communicate through end-to-end mTLS connections. The mirrored network pattern prevents direct communication between environments unless it is explicitly required and controlled. When protocols and authentication differ, an API gateway or proxy such as Apigee acts as a unifying facade, and private paths can use Direct VPC egress from Cloud Run into the cluster VPC through an internal passthrough Network Load Balancer.
GKE and Cloud Run are two container options that differ in control, workload fit, and billing. GKE gives full control over Kubernetes clusters, while Cloud Run runs containers directly on Google Cloud infrastructure with limited infrastructure control:
| Feature | GKE | Cloud Run |
|---|---|---|
| Deployment and management | Manage clusters, including nodes, networking, scaling, and upgrades | Run containers from source code or an image without managing clusters |
| Control and flexibility | Full control and advanced customization | Limited control over environment variables, concurrency, and networking |
| Application types | Stateless and stateful, including complex apps with specific resource needs | Best for stateless, request- or event-driven web services and functions |
| Pricing model | Pay per cluster per hour, in Standard or Autopilot mode | Pay per use, rounded up to the nearest 100 milliseconds |
GKE Enterprise manages these workloads as a fleet, which is a logical grouping of clusters that share security, configuration, and service management. Fleet tooling includes Config Sync, Policy Controller, Config Controller for declarative Google Cloud resources, and Cloud Service Mesh for on-premises or cloud services. Centralized logging and monitoring give one view of metrics in the Google Cloud console, and Cloud Deploy with Cloud Build and Artifact Registry provides consistent delivery to GKE, Cloud Run, or both.
Strategic motivations for hybrid and multi-cloud are the business reasons for running parts of an application in more than one environment. A hybrid architecture runs some parts in Google Cloud and other parts on-premises, while a multicloud architecture runs some parts in Google Cloud and other parts in other clouds. Common motivations are reducing vendor lock-in, choosing best-of-breed services, improving resilience, meeting compliance needs, and optimizing costs. Operational complexity is higher because security tools are not standard across platforms and provisioning takes more effort.
Exam tip: Hybrid connects Google Cloud with on-premises, while multicloud connects Google Cloud with other cloud providers.
Resilience and disaster recovery use a second environment as a standby so work continues if the primary fails. In a hybrid design, a mission-critical on-premises application backs up data to Google Cloud and activates a replica there if on-premises goes down. In a multicloud design, Google Cloud can be primary and another cloud can hold a passive replica that restores production if Google Cloud fails. Backup frequency and whether the replica is active or passive depend on the recovery time objective (RTO) and recovery point objective (RPO).
Access to best-of-breed capabilities means using the strongest service from each provider without moving everything. Organizations often want Google Cloud strengths such as low-cost archive storage, artificial intelligence (AI) and machine learning (ML), and BigQuery for warehousing and analytics. In a hybrid pattern, data flows from the on-premises application to Google Cloud for analysis. In a multicloud pattern, data flows from applications on other clouds into Google Cloud for processing.
Development control and tiered topologies describe how hybrid architectures split where code is built and where it runs. Development environments stay on-premises, and a CI/CD pipeline pushes updates from on-premises development to Google Cloud production. This gives control over development while gaining cloud scalability and reliability. A tiered topology, also called split-stack deployment, places the application frontend in Google Cloud and the backend on-premises. Cloud bursting sends overflow traffic to Google Cloud when load on the on-premises application reaches peak capacity, then shrinks the extra capacity when demand falls.
Compliance and monitoring patterns separate sensitive data and centralize visibility. The separate application and operations pattern segregates sensitive application data from less sensitive operations data to meet compliance rules. The single pane of glass pattern centralizes monitoring and logging so data flows from on-premises and cloud sources into one cloud-hosted repository with one view, one place for retention, and centralized access control and auditing.
Compliance and data sovereignty are legal rules that control where data may be stored and how sensitive systems connect to the cloud. Data residency rules can require PII to stay inside a specific country, and a common compliant solution is to use another CSP with a local region when the current CSP has none there. Hybrid designs keep highly secure data and legacy systems on-premises and expose only needed capabilities to cloud applications. A common pattern is to modernize legacy services with API interfaces and manage them with Apigee for security, analytics, and scalability with minimal changes.
Secure integration prevents uncontrolled traffic between sensitive on-premises systems and public cloud. The mirrored network pattern ensures systems in different environments do not communicate directly, and any required cross-environment communication happens in a controlled manner. A zero trust service mesh helps maintain that separation, and connectivity is established based on the selected network design and landing zone hierarchy for organization, projects, and resources.
Google Kubernetes Engine (GKE) (opens in a new tab) simplifies portable workloads for disaster recovery, bursting, and geographic distribution. Disaster recovery keeps a standby system with the same application versions and configurations as the primary, so standby deployment is integrated into the CI/CD pipeline and artifact repositories are designed to avoid a single point of failure. Traffic shifts with DNS and load balancing, and key practices include:
Handling peak demand with cloud bursting runs the base workload in one environment and overflows extra demand to the public cloud. Bursting works only when environments are consistent and connected, with portable code, aligned CI/CD versions, high-volume links, and load balancers splitting traffic. Reducing latency through geographic distribution places services close to users, downstream services, or large data sets that cannot move. Combining multiple regional load balancers with DNS routing policies directs each client to the closest healthy environment, which also provides higher availability when one location has problems.
System architecture showing on-premises base workload handling normal traffic and bursting overflow to an autoscaled GKE cluster in Google Cloud via Cloud Load Balancing, Cloud DNS and Cloud Interconnect with a shared container image.
Hybrid and multicloud motivations differ mainly in what runs where and why that split helps the business. The comparison shows that hybrid links on-premises systems to the cloud, while multicloud combines Google Cloud with other providers:
| Strategy | What runs where | Common motivations | Example use cases |
|---|---|---|---|
| Hybrid | Some parts in Google Cloud, other parts on-premises | Disaster recovery, development control, enhancing on-premises apps, split-stack tiers | On-premises app with Google Cloud recovery site; on-premises development with CI/CD to cloud; BigQuery and AI/ML for on-premises data; frontend in cloud with backend on-premises |
| Multicloud | Some parts in Google Cloud, other parts in other clouds | Disaster recovery, enhancing apps on other clouds, best-of-breed services | Google Cloud primary with another cloud as recovery site; archive, AI/ML, and BigQuery for apps on other clouds |
Cost and operational tradeoffs apply to both strategies. Running in more than one environment often costs more than running entirely in Google Cloud because data may be stored redundantly, transfer costs apply, redundant resources sit underutilized, and cross-cloud networking and multi-environment monitoring add expense. Organizations should compare cost per unit of work, such as users served or requests processed, and weigh competitive pricing against personnel costs and design overhead. Terraform can reduce provisioning effort, and GKE attached clusters can manage containerized applications across clouds.
GKE Enterprise is a single control panel for hybrid and multicloud infrastructure that enables consistent deployment and portable applications. Workload portability is the ability to move an application across computing environments with minimal changes, while environment consistency means those environments share the same runtime, configuration, security, and service management. Together they improve agility and reduce friction when workloads move between on-premises, edge, and cloud platforms.
Google Kubernetes Engine (GKE) provides the uniform runtime layer behind this model. Applications are packaged into containers, deployed as Pods to nodes, and managed through the Kubernetes API, where nodes are Compute Engine virtual machines grouped into a cluster. Kubernetes hides differences between edge locations and the cloud so teams develop and operate workloads the same way. GKE automatically upgrades the control plane and worker nodes for features and security patches, with release channels, maintenance windows, and exclusions controlling timing, and Autopilot mode lets Google Cloud manage worker nodes as well.
A fleet is a logical grouping of Kubernetes clusters that share consistent security, configuration, and service management. Fleet capabilities include Config Sync for deploying configuration from a source of truth, Connect gateway for multi-cluster access, Policy Controller for enforcement, and Cloud Build, Cloud Deploy, and Artifact Registry for integrated CI/CD. Using the same deployment, monitoring, and tooling in every environment reduces friction when moving workloads.
Workload portability considerations are the environment differences that can change how a moved application behaves. Kubernetes and GKE make a workload portable with fewer changes, but portability does not guarantee equal performance. Data movement is central to planning because moving a workload often requires moving its data. The main factors that affect portability are:
Teams should identify business use cases and the data that powers them, then consider storage type, capacity, accessibility, and movement options. Sensitive Data Protection helps discover, classify, and protect data assets for regulated industries with storage and cross-region limits. If workloads still run in virtual machines, Migrate for GKE can containerize them without rewriting applications.
Unified cluster management means operating clusters in different locations together as a fleet from Google Cloud. The fleet can include GKE clusters on Google Cloud, Google Distributed Cloud clusters on on-premises hardware or VMware vSphere, GKE on AWS and GKE on Azure infrastructure, and attached standard CNCF-compliant clusters. An operator connects through one management layer, including the Connect gateway, instead of using different tools per environment. Google Cloud manages the control plane and system components, so platform teams do not maintain the Kubernetes API server themselves.
Shared tooling applies the same configuration, health management, and observability to every cluster. Configuration flows from a source of truth to each cluster, logging and monitoring are built in with Google Cloud Observability, and upgrades and repairs are automated with release channels, maintenance windows and exclusions, node auto-repair, and automatic scaling based on Pod count. Integrated CI/CD with Cloud Build, Cloud Deploy, and Artifact Registry moves build and deployment work into managed services for security, scale, and simplicity.
Cost-effective scaling means paying closely for compute that workloads actually use while nodes scale automatically. GKE supports automatic scaling of nodes based on Pod count, with a free tier to start without cluster management cost. The choice between modes affects management and billing:
| Mode | Who manages infrastructure | What determines compute cost |
|---|---|---|
| Autopilot mode | Google Cloud manages control plane and worker nodes | Pay only for compute resources that running Pods request |
| Standard mode | Google Cloud manages control plane while team manages nodes and pools | Pay for all resources on nodes, regardless of Pod requests |
Autopilot minimizes overhead through hardening, best-practice configuration, and Google-managed upgrades, with greater than 99% monthly uptime SLO and Pod-level SLA. Fault-tolerant workloads such as batch jobs can run on Spot Pods at lower cost.
Centralized governance means one control point sets and enforces the same security, compliance, and configuration rules everywhere. A single GKE control plane includes the API server, cluster state database built on etcd or Spanner, kube-controller-manager, kube-scheduler, and control plane virtual machines (VMs). Google manages these components inside a Google-owned project that customers cannot directly modify, while customers remain responsible for worker nodes and workloads. Security fixes and configuration therefore flow from Google to every connected environment the same way.
Standardized security uses Center for Internet Security (CIS) benchmarks with different scopes:
| Benchmark | Applies to | Use it for |
|---|---|---|
| CIS Kubernetes Benchmark | Open source Kubernetes project | Guidance for many self-managed and hosted implementations |
| CIS GKE Benchmark | Components customers control in GKE on Google Cloud | GKE-specific recommendations for nodes and workloads |
Teams should prioritize the CIS GKE Benchmark for their own responsibilities, plus operating system guidance such as Container-Optimized OS Benchmark, Ubuntu Linux Benchmark for Standard clusters using Ubuntu, and Windows Server Benchmark for Standard clusters using Windows. Consistent configuration uses Pod Security Admission in all editions and optional Policy Controller enforcement of Pod Security Standards. Identity Service Config connects clusters to external identity providers, Authenticator Groups Config connects authentication to Google Groups for RBAC, and Secret Manager Config centralizes secrets. Cloud Service Mesh (opens in a new tab) then monitors and manages service communication, while zero trust mTLS, VPN or TLS encryption, and common identity secure traffic across environment boundaries.
Exam tip: Choose Autopilot mode to pay only for Pod requests with fully managed nodes, and choose Standard mode when the team needs to manage node configuration and pays for full node capacity.
A hybrid architecture combines on-premises infrastructure with public cloud, while a multi-cloud architecture uses services from more than one cloud service provider (CSP). Organizations choose these models for flexibility, compliance, and resilience, with the tradeoff of added complexity and operating cost compared with running in one cloud.
Use Google Kubernetes Engine (GKE) for stateful, complex, or customized workloads that need full control over Kubernetes clusters, including nodes, networking, scaling, and upgrades with pay-per-cluster pricing. Use Cloud Run for stateless, request- or event-driven web services and functions that run containers without managing clusters with pay-per-use billing rounded up to the nearest 100 milliseconds.
Google Kubernetes Engine (GKE) Enterprise operates Kubernetes clusters across Google Cloud, on-premises, edge, and other clouds as a fleet with shared security, configuration, and service management. A fleet is a logical grouping of clusters that uses tooling such as Config Sync, Policy Controller, and Cloud Service Mesh to apply consistent deployment, policy, and observability in every environment.
In Autopilot mode Google Cloud manages the control plane and worker nodes and charges only for compute resources that running Pods request. In Standard mode Google Cloud manages the control plane while the team manages nodes and pools and pays for all resources on nodes regardless of Pod requests.
Last updated