Prepare and test your skills

Prepare and test your skills

Scalability and performance continuity mean an application keeps serving users as demand changes and as failures or stress occur. In Google Cloud, these goals are met by combining failure domains, location-scoped resources, managed services, redundancy, load balancing, and disaster recovery planning. Scalable and resilient applications (opens in a new tab) use enough capacity and keep that capacity available when conditions change.
Reliability means an application meets its current objectives for availability and resilience to failures. Availability is the percentage of time an application is usable, often described as nines. For example, 99.99% availability allows no more than 8.64 seconds of downtime in a 24-hour period, or no more than 10 failed requests out of 100,000. Different workloads measure reliability differently: content applications care about availability, latency, and throughput; databases and storage care about latency, throughput, availability, and durability; analytics pipelines care about consistent throughput and latency; and most applications care about data correctness. Application reliability depends on internal design, dependent components, Google Cloud infrastructure resources, provisioned capacity and how it scales, and DevOps processes and tools.
A failure domain is a resource or group of resources that can fail independently of other resources. A standalone Compute Engine virtual machine (VM) is a failure domain, while a Google Cloud region or zone is a failure domain made of a group of resources. Regions are independent geographic areas that consist of zones, and zones and regions are logical abstractions of underlying physical resources. Distributing an application redundantly across failure domains can produce higher aggregate availability than any single failure domain provides. Building blocks of reliability (opens in a new tab) explain how zonal, regional, multi-region, and global scopes affect resilience:
| Resource scope | Resilience | Examples | Mitigation |
|---|---|---|---|
| Zonal | Low | Compute Engine VMs, Hyperdisk volumes, single-zone GKE clusters, Filestore Basic and Zonal instances, Dataflow jobs, Cloud SQL instances | Deploy resources redundantly in multiple zones or regions. |
| Regional | Medium | Regional MIGs, regional Cloud Storage buckets, regional GKE clusters, VPC subnets, regional external Application Load Balancers, regional Spanner instances, Filestore Enterprise instances, Cloud Run services | Deploy resources redundantly in multiple regions. |
| Multi-region | High | Dual-region and multi-region Cloud Storage buckets, multi-region Spanner instances, multi-region Bigtable instances, multi-region key rings | Manage changes carefully, and use defense-in-depth fallbacks. |
| Global | High | Projects, VPC networks, Cloud DNS zones, global external Application Load Balancers, Pub/Sub topics, Secret Manager secrets | Manage changes carefully, and use defense-in-depth fallbacks. |
Zonal resources can be affected by a failure in their zone, but a zone failure usually does not affect other zones in the region. Regional resources are deployed redundantly across multiple zones and are resilient to zone incidents, though a region outage can affect them. Multi-region resources are distributed across specific regions and global resources are available across all Google Cloud locations, so both are resilient to zonal and regional incidents. Global resources are easier to set up as single instances, but a critical configuration error can make a global resource a single point of failure, so changes need careful control and regional defense-in-depth fallbacks can help.
Google Cloud infrastructure is designed to support target availability levels for most workloads, but specific service uptime service level agreements (SLAs) can differ from those targets. Architecting disaster recovery (opens in a new tab) uses deployment location as a guide to expected availability:
| Deployment location | Availability | Estimated maximum downtime |
|---|---|---|
| Single zone | 3 nines: 99.9% | 43.2 minutes in a 30-day month |
| Multiple zones in a region | 4 nines: 99.99% | 4.3 minutes in a 30-day month |
| Multiple regions | 5 nines: 99.999% | 26 seconds in a 30-day month |
A service actual SLA can depend on configuration. For example, a Bigtable instance with clusters in three or more regions has a minimum uptime SLA of 99.999% only if multi-cluster routing is configured; with single-cluster routing, the minimum is 99.9% regardless of cluster count. A multi-cluster Bigtable instance in multiple zones within a single region with multi-cluster routing has a minimum uptime SLA of 99.99%. Aggregate availability is the overall availability of a multi-tier stack and is affected by every tier. It is calculated by multiplying tier values: tier1_availability x tier2_availability x tierN_availability. For example, four services each at 99.9% produce 99.6% aggregate availability.
Side-by-side comparison showing chained services multiply to 98 percent and lower availability, while parallel services use 1 minus failure product to reach 99.99 percent and higher availability.
| Approach | Formula | Example with two 99% services | Effect |
|---|---|---|---|
| Chained services | A x B | 98% | Lower than each individual service. |
| Parallelized services | 1 - (1 - A) x (1 - B) | 99.99% | Higher than each service because services run independently. |
Where feasible, choose parallelization over chaining because chaining makes end-to-end availability depend on every service in the chain. Adding tiers to a chained design reduces aggregate availability, since the aggregate is generally lower than the least available tier.
Managed services help maintain performance continuity because they provide built-in features for backups, high availability, and scalability. For data management, use services such as Cloud SQL, Cloud Storage, and Spanner. For compute and application hosting, use Compute Engine managed instance groups (MIGs) and Google Kubernetes Engine (GKE) clusters; regional MIGs and regional GKE clusters are resilient to zone outages. To improve resilience against region outages, use managed multi-regional services. Traffic and capacity practices that support continuity include:
High availability (HA) and disaster recovery (DR) are different concepts with different scopes and replication methods. HA generally applies to regional deployments, while DR generally applies to multi-region deployments. Many managed services provide synchronous replication between zones within a single region by default. DR across two or more regions requires a defined replication strategy if multi-regional or global services are not used, and that replication is typically asynchronous.
| Concept | Scope | Replication | Recovery goals and topology |
|---|---|---|---|
| High availability (HA) | Regional deployments | Synchronous replication between zones, often by default | Supports zero or near-zero recovery time objective (RTO) and recovery point objective (RPO), and can create an active-active topology with no single point of failure. |
| Disaster recovery (DR) | Multi-region deployments | Typically asynchronous when multi-regional or global services are not used | Requires assessment of RTO and RPO effects and identification of manual or semi-automated failover steps. |
For financial institutions, failover region choice can be limited by data sovereignty and residency rules. Reliability for financial services (opens in a new tab) recommends active-active topology across two regions with managed multi-regional services such as Spanner and Cloud Storage when replication is critical. For critical financial applications, a multi-region topology across at least two regions and three zones in each region is recommended, because failure in one zone or region can be followed by disruption in a second zone.
Exam tip: HA applies to regional deployments and commonly uses synchronous replication between zones, while DR applies to multi-region deployments and typically uses asynchronous replication.
Change management protects continuity by reducing the reliability risk of infrastructure changes. Deploy infrastructure changes progressively: change a few VMs in one zone, monitor the effects, revert quickly if issues appear, diagnose and resolve the issues, then restart and gradually deploy across all infrastructure. Control changes to global resources with extra care because they can become single points of failure; classify modifications to global resources as high-risk changes that additional reviewers must verify. Monitoring and testing confirm that continuity strategies work. Use the Google Cloud Service Health Dashboard to view current service status across regions, and use its history page to see incident duration, affected zones and regions, affected services, and workarounds. Personalized Service Health shows incidents relevant to a project, supports API requests, and lets you configure alerts. Future directions for infrastructure reliability (opens in a new tab) include rigorous testing in a dedicated production-equivalent environment where network connectivity is physically disconnected to validate behavior during total zonal or regional failure. Customers can use Fault Injection Testing and simulated failure scenarios to test their own recovery.
Disaster recovery (DR) is a subset of business continuity planning (BCP). DR focuses on recovering from disruptions and resuming normal operations with minimal impact. DR planning begins with a business impact analysis that defines recovery time objective (RTO) and recovery point objective (RPO). RTO is the maximum acceptable time an application can be offline, while RPO is the maximum acceptable time during which data might be lost.
Recovery objectives define how much downtime and data loss a business can tolerate. RTO measures acceptable downtime, and RPO measures acceptable data loss in time, not the amount or quality of lost data. RTO and RPO are measurable and should be treated as service level objectives (SLOs). A service level agreement (SLA) is the full agreement covering service, support, costs, performance, penalties, and responsibilities, while SLOs are specific measurable parts of that agreement. Smaller RTO and RPO values usually cost more to run and add administrative complexity. For example, 99.9% availability allows about 86 seconds of downtime per day, while 99% allows 864 seconds, which is 10 times more downtime. Disaster recovery planning (opens in a new tab) starts from these objectives so architecture and budget choices are clear.
High availability (HA) and disaster recovery (DR) are related but different. HA generally applies to regional deployments, while DR generally applies to multi-region deployments. Many managed services provide synchronous replication between zones within a single region by default, supporting zero or near-zero RTO and RPO and enabling an active-active topology without a single point of failure (SPOF). DR across two or more regions requires a replication strategy if multi-regional or global services are not used, and that replication is typically asynchronous. Teams must assess how replication affects RTO and RPO and identify manual or semi-automated failover operations. Data sovereignty and residency rules can limit which failover region is allowed. For active-active operation across two regions, managed multi-regional services such as Spanner and Cloud Storage are recommended when data replication is critical.
Exam tip: RTO is the maximum acceptable downtime, while RPO is the maximum acceptable data loss measured in time. HA usually uses synchronous replication within a region, while DR usually spans regions and often uses asynchronous replication.
Backup and replication strategies determine how data and applications are copied and restored after a disruption. Use managed multi-regional storage services for data, and take snapshots of persistent disks and store them in multi-region locations. When using regional or zonal resources, set up data replication to other regions. Validate DR plans regularly, and track RTO, RPO, and the impact tolerance required by regulations. Managed services provide HA backed by SLAs, built-in backup mechanisms, and resilience features. For data management, consider Cloud SQL, Cloud Storage, and Spanner. For compute and application hosting, consider managed instance groups (MIGs) and Google Kubernetes Engine (GKE) clusters, because regional MIGs and regional GKE clusters are resilient to zone outages. Multi-regional services improve resilience against region outages, and services with unique characteristics may need an exit plan.
Location scope describes how widely a Google Cloud resource is distributed and how resilient it is to zone or region failures. Wider location scope generally improves resilience, but global resources still need careful change management because a configuration error can create a SPOF. Geography and regions (opens in a new tab) define where zonal, regional, multi-region, and global resources run:
| Resource scope | Examples | Resilience | Mitigation |
|---|---|---|---|
| Zonal | Compute Engine VMs, single-zone GKE clusters, Filestore Basic and Zonal instances | Low | Deploy resources redundantly in multiple zones or regions. |
| Regional | Regional MIGs, regional Cloud Storage buckets, regional GKE clusters, Filestore Enterprise instances | Medium | Deploy resources redundantly in multiple regions. |
| Multi-region | Dual-region and multi-region Cloud Storage buckets, multi-region Spanner instances | High | Manage changes carefully and use defense-in-depth fallbacks. |
| Global | VPC networks, Cloud DNS zones, global external Application Load Balancers | High | Manage changes carefully and use defense-in-depth fallbacks. |
Zonal resources can be affected by a zone failure, while regional resources are resilient to zone incidents but can be affected by a region outage. Multi-region and global resources are resilient to zone and region incidents, but a global configuration error can still make them unavailable everywhere. For exceptionally high availability, regional resources can act as defense-in-depth fallbacks. For example, a global load balancer can be the preferred frontend, while regional load balancers serve as backups that clients fail over to if the global load balancer is unavailable. Automated failover can use Cloud DNS (opens in a new tab) and regional backup load balancers.
Fault tolerance and aggregate availability describe how well a system keeps working when individual components fail. To eliminate SPOFs, distribute resources across locations and use redundant resources. Avoid deploying only a single application server or database. Use MIGs to recreate failed VMs automatically, use Cloud Load Balancing to distribute traffic evenly, use HA configurations for databases such as Cloud SQL, and use regional persistent disks with synchronous replication to improve data availability. The aggregate availability of a multi-tier stack is calculated as tier1_availability x tier2_availability x tierN_availability, so adding tiers usually lowers overall availability. For example, four services at 99.9% each produce an aggregate of 99.6%. Where possible, choose parallelization over chaining: two chained services at 99% produce 98% aggregate availability, while parallelized services produce 99.99% using 1 - (1 - A) x (1 - B). For critical financial services, use a multi-region topology across at least two regions and three zones in each region, overprovision capacity by one-third more than estimated, and minimize manual recovery with active-active designs. Reliability for financial workloads (opens in a new tab) ties these choices to regulatory impact tolerance.
Cloud Storage and Filestore are Google Cloud storage services that support recovery and continuity planning. Cloud Storage is a highly durable object store with redundancy across regions, storage classes, object lifecycle management, data transfer from other sources, encryption at rest by default, and soft deletion. Dual-region and multi-region Cloud Storage buckets are multi-region resources, so they are resilient to zone and region incidents. Filestore Basic and Zonal instances are zonal resources, while Filestore Enterprise instances are regional resources. Regional Filestore Enterprise is resilient to zone outages, but a region outage can affect it, so cross-region replication or another recovery strategy is needed. Persistent disk snapshots can be stored in multi-region locations, and regional persistent disks provide synchronous replication. Recovery scenarios for applications (opens in a new tab) show how these storage choices combine with replication to meet RTO and RPO goals.
Testing and validating continuity plans confirms that DR plans work before a real disruption occurs. DR plans should be tested regularly and should cover zonal and regional outage scenarios. Google Cloud conducts rigorous resilience testing in a dedicated, production-equivalent environment using an internal-only region that is architecturally identical to customer-facing regions. During this testing, network connectivity is physically disconnected to validate how products behave during a total zonal or regional failure. Local services in the affected area might go offline as expected, while control planes and services in other zones and regions remain unaffected. Google Cloud also runs annual company-wide Disaster Recovery Testing (DiRT) events, and customers can use Fault Injection Testing and simulated failure scenarios to test their own services.
Resilience and fault tolerance in Google Cloud mean designing infrastructure so work continues when parts fail. A resilient application keeps serving users despite zone or region failures, while a fault-tolerant design uses redundant copies that take over automatically. High availability, the share of time a service is usable, is increased by distributing compute, balancing traffic, and replicating data.
A single point of failure (SPOF) is any critical component that depends on one infrastructure resource, so failure of that resource makes the whole stack unavailable. A typical three-tier application has a web tier with a load balancer and web servers, an app tier with application servers, and a data tier with a database. If the app tier runs on a single virtual machine (VM), or if there is only one load balancer, one app server, or one database, failure of any of those single resources causes user requests to fail. Removing a SPOF requires distributing resources across locations and deploying redundant resources. Every critical component must be replicated across multiple machines, zones, and regions so no sole zone or region can cause a global outage. Automatic failover must then shift work to the healthy copy when a zone or region has an outage. Designing reliable infrastructure (opens in a new tab) centers on this removal of single dependencies.
Distributing workloads means choosing a location scope that limits how far a failure can spread. Compute Engine VMs are zonal resources that live in one zone, while regional resources such as regional managed instance groups (MIGs), regional subnets, and regional persistent disks are spread redundantly across multiple zones in a region. Multi-region resources such as multi-region Cloud Storage buckets and multi-region Spanner instances are spread across regions, and global resources such as Virtual Private Cloud (VPC) networks and global external Application Load Balancers do not depend on any single region. Geography and regions (opens in a new tab) define these boundaries and how failures are contained:
| Architecture | Target availability | Use when |
|---|---|---|
| Single-zone deployment | 99.9% | The workload can tolerate downtime or can be redeployed elsewhere with minimal effort |
| Multi-zone deployment | 99.99% | The workload needs resilience to zone outages but can tolerate downtime from a region outage |
| Multi-region deployment | 99.999% | The workload is business-critical and high availability is essential, such as retail and social media applications |
A zone defines the expected boundary for a failure, and a failure in one zone usually does not affect resources in other zones in the region. Regional resources survive a single-zone incident but can be affected by a region outage from natural disasters or large infrastructure failures. Multi-region and global resources survive single-region outages, so they provide the highest resilience.
Managed services remove single points of failure by building redundancy, health checking, and failover into the service itself. Cloud Load Balancing, through a global external Application Load Balancer, receives client traffic and forwards requests only to healthy backends, and it is resilient to zone and region outages. Cloud SQL high availability (HA) provisions a primary-standby pair of database instances in separate zones with synchronous replication, so the Cloud SQL service fails over to the standby database automatically if the primary zone fails. A resilient multi-tier topology combines zonal, regional, and global resources so each tier has a surviving copy. Traffic flows through an ordered path:
If a VM crashes, the load balancer sends traffic to the other VMs while the MIG replaces the failed VM. If the primary database zone fails, traffic continues through the standby database, and if a whole region fails, the database can be restored in a different region from the multi-region backups. Patterns for scalable and resilient apps (opens in a new tab) describe this ordered flow from frontend to compute to data.
Exam tip: A global load balancer is resilient to zone and region outages, but a configuration error can still make it a single point of failure across all geographies, so manage changes carefully.
A distributed architecture with redundant resources costs more and is more complex to operate than a single-zone design. More resources must be provisioned and managed, more cross-location network traffic flows between zones and regions, and more data is stored and replicated. Latency and operational work can also increase as requests and data move across locations. Building blocks of reliability (opens in a new tab) frame this as a tradeoff between wider scope and higher management cost. Choose redundancy based on business need rather than using multi-region for everything. For business-critical applications, the availability gain often outweighs the higher cost and complexity, while workloads that tolerate downtime can stay single-zone.
High availability (HA) generally applies to regional deployments with synchronous replication between zones within a single region, supporting zero or near-zero recovery time objective (RTO) and recovery point objective (RPO) and enabling an active-active topology. Disaster recovery (DR) generally applies to multi-region deployments and typically uses asynchronous replication when multi-regional or global services are not used, requiring assessment of RTO and RPO effects and identification of manual or semi-automated failover steps.
Recovery time objective (RTO) is the maximum acceptable time an application can be offline, while recovery point objective (RPO) is the maximum acceptable time during which data might be lost. Smaller RTO and RPO values usually cost more to run and add administrative complexity.
Aggregate availability for chained tiers is calculated by multiplying tier values as tier1_availability x tier2_availability x tierN_availability, so four services each at 99.9% produce 99.6% aggregate availability. Parallelized services use 1 - (1 - A) x (1 - B) and achieve higher availability, so two 99% services chained produce 98% while parallelized they produce 99.99%.
An organization is designing a disaster recovery (DR) strategy for a non-critical reporting database and needs to align its architecture with the following business requirements:
Which disaster recovery strategy should the organization implement?
Maintain a pre-provisioned standby database instance with continuous asynchronous replication in a secondary region
Configure an active availability group with synchronous replication across multiple regions
Deploy a multi-zone availability group within a single region using local disk snapshots
Perform scheduled database backups directly to a multi-region Cloud Storage bucket
Last updated