Deploy Scalable and Highly Available Databases in Google Cloud practice questions

Read replica in same zone

Why wrong: A read replica is a separate instance that replicates asynchronously. It does not provide automatic failover.

Zonal availability configuration

Why wrong: Zonal availability means a single-zone instance with no HA. Failure of the zone leads to downtime.

Cross-region failover replica

Why wrong: Cloud SQL does not support automatic failover across regions. Cross-region read replicas can be manually promoted but failover is not automatic.

Enable IAM database authentication

Why wrong: IAM authentication does not enforce SSL. It uses OAuth tokens but connections can still be plaintext.

Use the Cloud SQL Auth Proxy to connect

Why wrong: The Auth Proxy uses SSL but does not enforce it at the database level; clients can still connect directly without SSL if the flag is not set.

Assign a private IP address to the instance

Why wrong: Private IP does not encrypt traffic; it only restricts network access.

Use autoscaling with a minimum of 1000 processing units, a maximum of 2000 processing units, and a target CPU utilization of 65%

Why wrong: Spanner uses high-priority CPU utilization, not total CPU utilization, for autoscaling decisions.

Use autoscaling with a minimum of 1 node, a maximum of 2 nodes, and a target CPU utilization of 65%

Why wrong: Processing units are used, not nodes. Node-based scaling is legacy and not integrated with processing units.

Manually increase processing units to 2000

Why wrong: Manual scaling requires constant monitoring and adjustment; autoscaling is more efficient.

Standard machine type with HDD and storage auto-increase enabled

Why wrong: Standard machine type may not provide enough memory for 30,000 TPS, and HDD is too slow for high transaction throughput.

Custom machine type with 4 vCPUs and 15 GB memory, SSD, and storage auto-increase enabled

Why wrong: 15 GB memory may be insufficient for 30,000 TPS heavy workloads; highmem typically offers 6.5 GB per vCPU.

Lightweight machine type with SSD and storage auto-increase disabled

Why wrong: Lightweight instances (shared-core) are not suitable for high-throughput production workloads. Disabling storage auto-increase risks downtime.

service cloud.firestore { match /databases/{database}/documents { match /{document=**} { allow read: if request.auth.uid == get(/databases/$(database)/documents/$(document)).data.userId; } } }

Why wrong: The get() function is unnecessary and slower. resource.data directly accesses the document data.

service cloud.firestore { match /databases/{database}/documents { match /{document=**} { allow read: if request.auth.uid == request.resource.data.userId; } } }

Why wrong: request.resource refers to the new data being written, not the existing document. For reads, use resource.data.

service cloud.firestore { match /databases/{database}/documents { match /{document=**} { allow read: if request.auth.uid == resource.data['user-id']; } } }

Why wrong: The field is 'userId', not 'user-id'. The rule would fail to match.

Production instance with a single node

Why wrong: Production instances incur full costs even with one node.

Production instance with a multi-cluster configuration

Why wrong: Multi-cluster configurations are for high availability and replication, not reduced cost.

Production instance with HDD storage

Why wrong: HDD is cheaper than SSD but still a production instance type, incurring costs.

Create a primary cluster in the primary region and a primary cluster in the secondary region, then set up bidirectional replication

Why wrong: AlloyDB does not support bidirectional replication; secondary clusters are read-only.

Use AlloyDB Omni in the secondary region to replicate data from the primary region

Why wrong: AlloyDB Omni is for on-premises deployments, not for cross-region replication within GCP.

Create a primary cluster with multiple read pools in the same region and distribute read traffic

Why wrong: This provides read scaling within the same region, not cross-region disaster recovery.

1000 processing units

Why wrong: 1000 PUs is 1 node.

4000 processing units

Why wrong: 4000 PUs equals 4 nodes.

500 processing units

Why wrong: 500 PUs is half a node, not 2 nodes.

The cbt command-line tool

Why wrong: cbt is for administrative tasks, not hot spot analysis.

Bigtable Studio

Why wrong: Bigtable Studio is not a real GCP tool.

Cloud Monitoring (formerly Stackdriver)

Why wrong: Cloud Monitoring provides metrics but not the specific hot spot visualisation that Key Visualiser offers.

Assign a private IP to the Cloud SQL instance and connect using it

Why wrong: Private IP alone does not provide authentication; you still need to configure IP allowlisting on the private IP range, which they want to avoid.

Use the Cloud SQL Connector library inside the VM

Why wrong: The Connector library requires the Auth Proxy or a direct connection; it is not a standalone connection method.

Use a direct IP connection with SSL enabled

Why wrong: Direct IP connection still requires public or private IP allowlisting, which they want to avoid.

Firestore

Why wrong: Firestore supports multi-region but is a document database with limited transaction support; primarily for mobile/web.

Cloud Bigtable

Why wrong: Bigtable is NoSQL and does not support ACID transactions.

Cloud SQL

Why wrong: Cloud SQL is regional and does not support multi-region deployment.

Create the replica in a different zone within the same region

Why wrong: Cross-region means different region, not different zone.

Configure an external replica in the target region

Why wrong: External replicas are for databases outside Cloud SQL, not for cross-region replication.

Enable HA on the primary instance first

Why wrong: HA is not a prerequisite for cross-region replicas.

Create a development instance type

Why wrong: Development instances cannot have multiple clusters.

Disable replication to avoid conflicts

Why wrong: Replication is needed for HA; conflicts are handled via last-write-wins.

Use HDD storage to reduce costs

Why wrong: HDD offers lower throughput and higher latency; not ideal for production.

Use nodes with a fixed number

Why wrong: Nodes are not compatible with autoscaling; processing units are used.

Use a regional configuration

Why wrong: Regional configuration covers only one region, not global.

allow create: if resource.data.owner == request.auth.uid;

Why wrong: For create, resource.data refers to the document if it exists, but create is on a new doc; request.resource is correct.

allow read: if request.resource.data.visibility == 'public';

Why wrong: For read, request.resource does not exist; use resource.data.

allow write: if request.auth.uid != null;

Why wrong: write includes create, update, delete; the requirement specifies only create and read.

Create a zonal Cloud SQL instance and set up a read replica in a different zone.

Why wrong: A zonal instance has no HA; a read replica is asynchronous and not used for automatic failover.

Create a Cloud SQL for PostgreSQL instance with cross-region replication.

Why wrong: Cross-region replication is asynchronous and does not provide automatic failover in under 60 seconds.

Configure a Cloud Spanner regional instance with multi-zone nodes.

Why wrong: Cloud Spanner is a different service; the question specifies migrating to Cloud SQL.

Firestore

Why wrong: Firestore is eventually consistent and limited in global distribution.

Cloud SQL

Why wrong: Cloud SQL is regional and does not provide global distribution.

Cloud Bigtable

Why wrong: Bigtable is eventually consistent across regions and not designed for OLTP with strong consistency.

Enable query distribution across read pool instances.

Why wrong: Query distribution is automatic but does not change the number of instances.

Configure a managed instance group for the read pool.

Why wrong: Managed instance groups are for Compute Engine, not AlloyDB.

Set up Cloud SQL read replicas with cross-region replication.

Why wrong: This is Cloud SQL, not AlloyDB, and read replicas are not automatically scaled.

Connect via Cloud Shell using the mysql client.

Why wrong: Cloud Shell is a temporary environment, not suitable for production applications.

Set up a Cloud SQL private IP connection.

Why wrong: Private IP requires VPC peering; the application still needs to be in the same VPC to connect without IP allowlisting.

Use SSL/TLS connection with the public IP address.

Why wrong: SSL alone still requires whitelisting the client IP.

Cloud Monitoring dashboards

Why wrong: Monitoring shows aggregate metrics but not per-key hot spots.

cbt command-line tool

Why wrong: cbt is for table operations, not visualization.

Cloud Logging

Why wrong: Logging records events but doesn't visualize hot spots.