Prepare and test your skills
Prepare and test your skills
Data Protection and Encryption Strategies cover how to place application and database (opens in a new tab) copies so data survives failures, stays consistent, and meets residency rules. A primary location handles regular traffic and a standby location provides extra capacity or recovery. Databases can be isolated in one location, replicated between locations, or synchronized between locations. The placement choice controls availability during outages, delay in seeing updates, and whether regulated data stays inside a required country or boundary.
Disaster recovery within and across clouds uses a primary and secondary region plus a primary and secondary cloud to handle different sizes of failure. A whole cloud is less likely to fail than a single region, but running in two clouds still protects against very large failures and helps meet business and industry regulations. To recover an application after a failure, either the secondary region or the secondary cloud is used, depending on how severe the outage is. This lets an enterprise choose the best recovery path instead of relying on one failover target.
Cloud bursting is a configuration where application client traffic scales across deployment locations when demand exceeds the primary location. The primary location handles regular traffic and the standby location provides additional capacity, with application service instances deployed in both locations. Cloud bursting is implemented across clouds or in hybrid clouds, where a public cloud adds elastic capacity to limited on-premises compute resources.
Database placement for bursting determines where data flows and what consistency risk is accepted. In primary-location deployment, the database exists only in the primary location, so a bursting application in the standby location accesses the database in the primary location. In primary-and-standby deployment, a database instance exists in both locations and each location's application accesses its local instance, with the copies kept transactionally synchronized or asynchronously synchronized. Asynchronous synchronization can have a delay, updates made in the standby location must propagate back to the primary location, and conflict resolution must be implemented if both locations allow concurrent updates. When data must stay within a country and one public cloud has only one region there, bursting into a region of a different public cloud in the same country keeps data inside the country while adding capacity.
Exam tip: When capacity is constrained but data must stay in-country, burst to another cloud's region in the same country; use primary-only database placement when the standby must read from the primary, and primary-and-standby placement with synchronization when each location needs local access.
Partitioned, replicated, and synchronized data are three ways to share data when different parts of an application run in different clouds. Best-in-class use places different parts in different clouds, such as business-logic processing in one cloud and analytics in another cloud. A distributed service is deployed and executed in two or more deployment locations, with either all service instances in all locations or only some services in each location based on hardware availability or expected limited load. Data in a transactionally synchronized database is consistent in all locations, so it is the best option when service instances run everywhere.
| Data approach | How data is shared | Consistency and write behavior |
|---|---|---|
| Partitioned data | Each application part has its own separate database partition with no direct connection between databases | The managing application writes shared data twice so it exists in both partitions |
| Asynchronously replicated database | Data or a subset is copied from one cloud to the other after the original write | There is a delay before the second cloud sees updates; concurrent changes in two locations require a conflict-resolution strategy |
| Transactionally synchronized database | Updates from each application are immediately available to both partitions as a single distributed database | Each update is transactionally consistent in both locations without delay or merge conflicts |
When an asynchronously replicated database is used across locations, the same data item can be modified concurrently in two places. A conflict-resolution strategy must then determine which conflicting change becomes the final consistent state.
Identity and Access Management (IAM) and resource governance is the practice of giving the right identity only the permissions it needs and keeping administrative control as resources grow. In Google Cloud, Cloud Identity authenticates users and can federate with an existing identity provider, while IAM authorizes what users, groups, and service accounts can do on each resource. Least privilege, isolated service accounts, and fine-grained data controls work together so people and workloads cannot reach data or infrastructure outside their job.
Least privilege for users means granting only the minimum necessary IAM permissions for a person's role and removing anything unused. An administrator can receive full access to an environment's bucket and cluster, while a regular user receives a limited role such as Composer User (opens in a new tab), which enables access only to the DAG UI and Airflow UI. Airflow UI Access Control further reduces visibility inside Airflow based on the user's Airflow role and can assign permissions for individual DAGs. Access that is not explicitly granted stays denied, so new users and new data assets start with no access until an owner authorizes them.
Regular audits keep least privilege working over time:
These checks prevent permissions from growing as people change roles and as new tables and buckets are added.
An environment service account is the workload identity that an environment uses to operate infrastructure and run automated work such as Airflow DAGs. Each environment should use its own dedicated service account with only the permissions needed to operate that environment and perform the operations defined in its DAGs. Because DAG code runs with the permissions of the environment service account, a broad service account directly expands what any DAG in that environment can do.
Exam tip: Do not run an environment with the Editor basic role or reuse Google default service accounts, because reducing permissions on a default account can affect other services in the project while an Editor account lets DAGs use broader permissions than intended.
Fine-grained access in BigQuery (opens in a new tab) combines table-level IAM with column-level policy tags to enforce least privilege on sensitive data. The cdmc sensitive data classification policy tag taxonomy defines four tags: Sensitive Personal Identifiable Information, Personal Identifiable Information, Sensitive Personal Information, and Personal Information. These tags map to sensitivity classifications derived from Sensitive Data Protection infoTypes such as age, date of birth, phone number, and email address.
| Access level | Control mechanism | Default behavior |
|---|---|---|
| Table | IAM binding on the dataset or table | Only creators, owners, and explicitly granted identities can read the table |
| Column | Policy tag bound to the sensitive column | Users have no access to tagged columns until granted the policy-tag reader role |
A scheduled detector maintains this control by checking for assets where a sensitive category has no related policy tag or where the category does not match the policy tag. The detection flow runs in order:
A four-step process flow showing the sensitive data detection loop: Cloud Scheduler triggers the detector, findings are published to Pub/Sub, Dataflow stores them in BigQuery, and a remediation tool fixes entitlements, with the loop repeating.
events table in BigQuery.This loop keeps sensitive columns covered even as new tables appear and classifications change.
Protection of secrets and operational storage prevents a user with storage access from acting as a privileged workload. Secret Manager (opens in a new tab) stores Airflow connections, API keys, passwords, and certificates, so sensitive values do not appear in DAG files or environment variables. Access to the environment's bucket is limited to trusted users because a user who can write to the bucket can influence what the environment service account executes on their behalf. The same boundary applies to backups and network paths. Permissions to create environment snapshots and to access the bucket where snapshots are stored are granted only to trusted users, since snapshots contain environment data and configuration. All external interfaces use encryption by default, and connections to external products and services must use encrypted communication such as SSL or TLS.
Network Security and Perimeter Defense is the practice of blocking unwanted traffic at the edge, isolating sensitive resources from exfiltration, and allowing only verified access to backends. In Google Cloud, Google Cloud Armor (opens in a new tab) filters incoming traffic before it reaches load-balanced backends, VPC Service Controls (opens in a new tab) creates isolation perimeters around managed services, and Identity-Aware Proxy (IAP) (opens in a new tab) verifies identity before access is granted. Cloud Next Generation Firewall (opens in a new tab) firewall rules and private connectivity complete the perimeter by controlling which traffic can enter or leave Virtual Private Cloud (VPC) networks.
Google Cloud Armor is an edge security service for applications exposed behind Cloud Load Balancing that provides Layer 3 through Layer 7 protection. Each Cloud Armor security policy is made up of a set of rules that include attributes from Layer 3 through Layer 7, and the policy reviews incoming requests for common web attacks or other Layer 7 attributes. Traffic flows from the internet client to Google Front Ends (GFEs) to the Cloud Armor policy to backend services or backend buckets, so malicious traffic is blocked before it reaches the backend runtimes. The same attachment to backend services also enables DDoS mitigation, Web Application Firewall (WAF) screening, IP and geo-based filtering, and rate limiting.
Google Cloud Armor and VPC firewall rules have different functions that work together at the perimeter. Cloud Armor acts on client traffic at GFEs, while VPC firewall rules allow or deny traffic to and from backend virtual machines (VMs). Traffic proxied through a global external Application Load Balancer or classic Application Load Balancer can only reach backend instances when ingress allow firewall rules permit traffic from the load balancer IP ranges and health-check systems to those backends.
| Control | Where it acts | What it controls |
|---|---|---|
| Google Cloud Armor | At GFEs, before the backend service | Client requests to the load balancer, filtered by IP, geo, and Layer 7 attributes |
| VPC firewall rules | At backend VMs and endpoints | Traffic to and from backends, including allowed traffic from GFEs and health checks |
When both controls are configured, only external traffic proxied through the load balancer with its associated security policy can reach the instances, and direct traffic to the backends is denied.
VPC Service Controls is an additional layer of protection that prevents data loss by creating isolation perimeters around Google Cloud resources, sensitive data, and networks. A service perimeter creates a security boundary that allows free communication within the perimeter but, by default, blocks communication across the perimeter to Google Cloud managed services. The perimeter works specifically with Google Cloud managed services and does not block access to third-party APIs or services on the internet. The service perimeter also limits the usefulness of compromised credentials because the perimeter blocks requests to restricted services that originate from attacker-controlled endpoints outside the environment.
A VPC Service Controls perimeter can be configured to control three types of communication:
VPC Service Controls does not require a VPC network, because access can be granted from external IP ranges or specific IAM principals through access levels. Access from the internet to managed resources inside a perimeter is denied by default, and requests are allowed only when they meet an ingress rule or access level based on source IP address, identity, or source Google Cloud project. To use the Google Cloud console with protected resources, an access level must allow one or more IPv4 and IPv6 ranges or specific user accounts. The boundary behavior is strict in both directions. A VM in a VPC network that is part of a perimeter can read from or write to a Cloud Storage bucket in the same perimeter, but a copy operation between two buckets succeeds only when both buckets are in the same perimeter. VPC Service Controls does not allow VMs in networks outside the perimeter to access buckets inside the perimeter without an ingress policy, and it does not allow a VM inside the perimeter to access buckets outside the perimeter. For hybrid environments, private communication can be extended over an authorized VPN or Cloud Interconnect, but the VPC network that contains the landing zone from on-premises must be part of the perimeter, and VMs with private IP addresses secured by the perimeter cannot access managed resources outside it.
Identity-Aware Proxy (IAP) is a zero-trust access service that verifies a user's identity and then determines whether that user can access an application or backend. IAP is attached to the backend services of a global external Application Load Balancer or classic Application Load Balancer, which is the same place where edge Cloud Armor security policies are attached. IAP TCP forwarding also replaces permissive Secure Shell (SSH) and Remote Desktop Protocol (RDP) access, because inbound firewall rules for SSH and RDP are set with IAP IP ranges as sources instead of the open internet. Permissive SSH and RDP rules allow brute-force attacks, so restricting inbound access to specific resources and routing admin traffic through IAP confines access to authenticated identities.
When Cloud Armor and IAP protect the same backend service, the evaluation order depends on the load balancer type. For a backend service of a global external Application Load Balancer, Cloud Armor evaluation happens first; if Cloud Armor blocks the request, IAP does not evaluate it, and if Cloud Armor allows it, IAP then evaluates identity. For a backend service of a classic Application Load Balancer, IAP evaluation happens first; if IAP authenticates the request, then Cloud Armor evaluates it, and if IAP authentication fails, Cloud Armor does not evaluate it.
Exam tip: Global external Application Load Balancer evaluates Cloud Armor first and then IAP, while classic Application Load Balancer evaluates IAP first and then Cloud Armor.
Cloud Next Generation Firewall firewall rules enforce network-level allow and deny decisions for traffic entering and leaving VPC networks. By default, all outbound access to the internet is allowed, which is a security risk, so a deny-by-default egress policy blocks outbound traffic and requires specific allow rules for only the known, necessary destinations. Inbound rules follow the same least-privilege pattern by restricting access to specific resources and ranges, including allowing only load balancer and health-check sources to reach backends and allowing only IAP sources to reach SSH and RDP ports.
Two foundational settings reduce the exposed surface before detailed rules are written. The compute.skipDefaultNetworkCreation organization constraint skips creation of the default auto-mode VPC network with its pre-populated IPv4 allow rules, which is not recommended for production. Private Google Access, when enabled on all subnets, lets VMs and services without external IP addresses reach Google Cloud services privately, so workloads do not need public addresses to use Google APIs.
Google Cloud Armor filters client traffic at Google Front Ends (GFEs) before it reaches backend services or backend buckets, while VPC firewall rules allow or deny traffic to and from backend virtual machines (VMs). Cloud Armor screens requests by IP, geo, and Layer 3 through Layer 7 attributes, and traffic proxied through a global external or classic Application Load Balancer can only reach backends when ingress allow firewall rules permit traffic from the load balancer and health-check sources.
Use primary-location deployment when the database exists only in the primary location so a bursting application in the standby location accesses the database in the primary location, and use primary-and-standby deployment when each location needs local access with copies kept transactionally synchronized or asynchronously synchronized. Asynchronous synchronization can have a delay, updates made in the standby location must propagate back to the primary location, and conflict resolution must be implemented if both locations allow concurrent updates.
A VPC Service Controls service perimeter allows free communication within the perimeter but by default blocks communication across the perimeter to Google Cloud managed services. A virtual machine in a perimeter can read from or write to a Cloud Storage bucket in the same perimeter, but a copy operation between two buckets succeeds only when both buckets are in the same perimeter, and access from the internet is denied by default unless allowed by an ingress rule or access level.
For a backend service of a global external Application Load Balancer, Google Cloud Armor evaluation happens first and Identity-Aware Proxy (IAP) then evaluates identity only if Cloud Armor allows the request, while for a classic Application Load Balancer IAP evaluates first and Cloud Armor evaluates only if IAP authentication succeeds. IAP verifies user identity before granting access to applications and is attached to the same backend services where edge Cloud Armor security policies are attached.
An enterprise is designing its Google Cloud resource hierarchy. The organization consists of multiple business units, including a Finance division that manages several sensitive workloads across multiple projects grouped under a Finance folder.
The cloud architecture must satisfy the following security and governance requirements:
Finance folder.Which combination of IAM and resource governance practices should the cloud architect recommend?
Assign the Folder Admin role (roles/resourcemanager.folderAdmin) to a Google Group on the Finance folder, and grant the Service Account User role (roles/iam.serviceAccountUser) to all developers at the organization level.
Assign the Organization Administrator role (roles/resourcemanager.organizationAdmin) to individual users on the Finance folder, and assign a single custom IAM role to all microservices across the organization.
Assign the Folder IAM Admin role (roles/resourcemanager.folderIamAdmin) to a Google Group on the Finance folder, and provision dedicated, separate service accounts with granular predefined roles for each microservice.
Assign the Owner basic role (roles/owner) to individual members of the Finance security team on each project in the Finance folder, and attach the default Compute Engine service account to all microservices.
Last updated