Task 2.2: Planning and implementing storage and data solutions.

• ACloud SQL for PostgreSQL with read replicas in each of the three regions

  Incorrect

  Cloud SQL takes all writes on one primary instance, and read replicas can lag behind it, so a read in another region can be stale.

• BBigtable with clusters in three regions and multi-cluster routing

  Incorrect

  Bigtable is a wide-column NoSQL store with eventually consistent replication between clusters, not a relational database.

• CSpanner with a multi-region instance configuration

  Correct

  Spanner is a relational database that scales horizontally and keeps strongly consistent, synchronously replicated data across regions in a multi-region configuration.

• DFirestore in a multi-region location

  Incorrect

  Firestore is a document database; it is not a relational SQL database for this kind of transactional schema.

Two requirements decide this one: relational transactions with strongly consistent reads everywhere, and writes that will outgrow one server. Spanner is built for both at once.

• AArchive storage

  Incorrect

  Archive is meant for data that you plan to access less than once a year, not every month.

• BNearline storage

  Correct

  Nearline suits data read on average once a month or less, such as files added continuously and read monthly for analysis; its 30-day minimum storage duration fits six-month retention.

• CStandard storage

  Incorrect

  Standard is for frequently accessed data or data kept only briefly, and these files are read once a month and kept for months.

• DColdline storage

  Incorrect

  Coldline is meant for data read at most once a quarter, and each of these files is read every month.

Match the class to how often each object is read: about monthly points to Nearline, quarterly to Coldline, less than yearly to Archive, and frequent access to Standard.

• ACloud SQL for MySQL, to store every reading in one table

  Incorrect

  Hundreds of thousands of writes per second from 2 million devices exceed what a single relational primary is built for.

• BFirestore, to keep one document per reading for the dashboard

  Incorrect

  Sustained ingestion at this rate, with scans by device and time, is the access pattern of a wide-column store rather than a document database.

• CMemorystore for Redis, as the system of record for readings

  Incorrect

  Readings must be kept for years as the system of record, which is not the role Memorystore is designed for.

• DBigtable, to store and serve the readings by device and time

  Correct

  Bigtable handles very high write throughput with low-latency reads by row key, which suits time-series data keyed by device and time.

• EBigQuery, to hold the history for analysts' SQL queries

  Correct

  BigQuery is a serverless data warehouse built for SQL analysis over large historical datasets.

Serving and analytics have different shapes. Bigtable gives fast key lookups at huge write rates; BigQuery gives SQL over everything. Many telemetry designs use both.

• AA Filestore instance with an NFS share mounted on all eight VMs

  Correct

  Filestore is a managed NFS file server that many VMs and GKE Pods can mount at once.

• BA Local SSD on each VM, synchronized with rsync every minute

  Incorrect

  Eight disks kept in step by a periodic copy are still eight file systems, and writes between copies conflict.

• CA Cloud Storage bucket mounted with Cloud Storage FUSE on each VM

  Incorrect

  Cloud Storage FUSE mounts a bucket as files, but it is not a fully POSIX-compliant NFS file system for concurrent writers.

• DA single Persistent Disk attached to all eight VMs in read-write mode

  Incorrect

  A Persistent Disk can be shared between VMs only in read-only mode; multi-writer sharing is a Hyperdisk feature that needs a clustered file system, not an NFS share.

When an application expects a shared network file system, use a managed file service rather than block disks or object storage. Filestore provides NFS file shares without a self-managed server.

• AA regional bucket with Object Versioning and soft delete enabled

  Incorrect

  Neither setting creates a copy outside the bucket's single region, which is what surviving a regional outage requires.

• BA multi-region bucket with the default replication setting

  Incorrect

  Default replication across regions is asynchronous with a longer target, so it does not promise the 15-minute objective.

• CA regional bucket plus a Storage Transfer Service job copying to a second region every hour

  Incorrect

  An hourly copy can lose up to an hour of uploads and puts the copy in a different bucket with a different name.

• DA dual-region bucket with turbo replication enabled on the bucket

  Correct

  Turbo replication on a dual-region bucket is designed to replicate newly written objects to both regions within a 15-minute recovery point objective.

Location type sets where Cloud Storage keeps copies. Dual-region and multi-region buckets are geo-redundant under one name, and turbo replication adds a 15-minute RPO, available only on dual-regions.

• AMemorystore for Valkey

  Correct

  Memorystore is a fully managed in-memory service suited to fast caching such as session data.

• BCloud SQL for MySQL

  Incorrect

  A relational database is slower for this read pattern than an in-memory store.

• CCloud Storage objects

  Incorrect

  Object storage is not designed for sub-millisecond reads on every request.

• DBigQuery tables

  Incorrect

  BigQuery is an analytics warehouse, not a low-latency session store.

Put hot, disposable data in memory. Managed in-memory stores give low latency without running cache servers yourself.

• AA dual-region bucket pairing europe-west3 with europe-west4

  Incorrect

  europe-west4 is in the Netherlands, so the second copy of the data would be stored outside Germany.

• BA bucket in the EU multi-region

  Incorrect

  A multi-region spans several regions, which may be outside Germany.

• CA regional bucket in europe-west3

  Correct

  A regional bucket stores its data in that one region.

• DA bucket in the US multi-region with a lifecycle rule

  Incorrect

  Lifecycle rules change classes or delete objects; they do not move data to another location.

Bucket location decides where data lives. When data must stay in one place, choose the region that matches the requirement.

• AA Compute Engine VM running sync scripts for both clouds

  Incorrect

  This works but means running and monitoring a VM and scripts.

• BA Filestore instance mounted on the S3 bucket

  Incorrect

  Filestore cannot mount an S3 bucket.

• CA scheduled Storage Transfer Service job from Amazon S3

  Correct

  Storage Transfer Service copies from Amazon S3 without agents and can run on a schedule.

• DDownloading the files to a laptop and uploading them again

  Incorrect

  80 TB through a laptop is slow and manual.

For cloud-to-cloud object migration, Storage Transfer Service is the managed, schedulable option.

• AA larger machine type for the instance

  Incorrect

  More capacity does nothing for a zone failure.

• BHigh availability on the instance

  Correct

  An HA instance keeps a standby in another zone and fails over automatically.

• CAutomated daily backups

  Incorrect

  Backups need a restore, which is neither automatic nor lossless.

• DA read replica in the same region

  Incorrect

  Read replicas serve reads; they do not fail over automatically for the primary.

High availability protects a Cloud SQL instance from zone failure. Backups and replicas solve other problems.

• AA Pub/Sub topic that the files are published to

  Incorrect

  Pub/Sub carries messages; it does not load files into BigQuery.

• BA Google-provided Dataflow template for text files

  Correct

  Google-provided templates run common pipelines on Dataflow, with a user-defined function for small transformations.

• CA custom Apache Beam pipeline written by the team

  Incorrect

  This works, but it is the code the team wants to avoid maintaining.

• DA Managed Airflow environment that copies the files

  Incorrect

  Managed Airflow orchestrates jobs; it is not the pipeline itself.

Templates package common Dataflow pipelines so teams configure them instead of writing them.