Analysing and Optimising Technical and Business Processes practice questions

Migrate for Compute Engine

Why wrong: Migrate for Compute Engine is for migrating virtual machines, not databases.

Cloud Dataflow

Why wrong: Cloud Dataflow is a data processing service, not a database migration tool.

Cloud Scheduler

Why wrong: Cloud Scheduler is for scheduling jobs, not for database migration.

Cloud SQL Auth Proxy

Why wrong: Cloud SQL Auth Proxy is used for secure client connections to Cloud SQL, but it is not a migration tool and is not required for the migration.

Transition to Coldline after 90 days, then to Archive after 1 year

Why wrong: Coldline is more expensive than Nearline for the first year; Nearline is sufficient for the 90-day requirement.

Transition to Archive after 90 days

Why wrong: Archive after 90 days is acceptable for cost but retrieval may exceed 1 hour (typically minutes). However, it's cheaper to use Nearline first to save on early access costs.

Transition to Standard after 90 days, then to Archive after 1 year

Why wrong: Standard is more expensive than Nearline; unnecessary cost for rarely accessed data.

Add a Cloud SQL Auth Proxy with connection pooling

Why wrong: Cloud SQL Auth Proxy does not provide connection pooling; it only secures connections. Connection pooling would require a separate proxy or application-level change, so this does not improve read throughput without code changes.

Use Cloud Memorystore for caching session data

Why wrong: Memorystore caches session data, which reduces database writes for sessions but does not directly improve read throughput for transactional queries.

Enable query caching in Cloud SQL

Why wrong: Query caching is deprecated in MySQL 8.0 and is not recommended; it may even cause performance issues.

Use SELECT * only when necessary

Why wrong: This is a best practice but not a specific action to take on the table structure.

Use materialized views to pre-aggregate data

Why wrong: This is also beneficial but not the primary two actions for reducing scans in this scenario.

Enable VPC Flow Logs to monitor egress.

Why wrong: VPC Flow Logs monitor traffic but do not reduce egress costs.

Use premium tier network for egress traffic.

Why wrong: Premium tier network improves latency but increases egress costs compared to standard tier.

Use a multi-regional Cloud Storage bucket to serve content.

Why wrong: Multi-regional Cloud Storage may increase egress costs due to higher per-GB charges.

Allow public access to the bucket and use signed URLs.

Why wrong: Public access with signed URLs does not reduce egress costs.

3-year committed use discount (CUD)

Why wrong: Similar to 1-year CUD, a 3-year commitment is too long and inflexible for a batch job that runs only 4 hours per day.

1-year committed use discount (CUD)

Why wrong: CUDs require a 1-year commitment, which is not flexible for a nightly 4-hour job. They also require consistent usage to be cost-effective.

Sustained use discounts

Why wrong: Sustained use discounts apply automatically but require running a VM for a significant portion of the month. A nightly 4-hour job may not benefit fully.

Datastream

Why wrong: Datastream is for streaming data to BigQuery, Cloud Storage, or Pub/Sub. It is not designed for database migration to Cloud SQL.

Migrate for Compute Engine (formerly Velostrata)

Why wrong: Migrate for Compute Engine is for migrating entire VMs, not databases specifically. It does not handle database schema conversion.

Transfer Appliance

Why wrong: Transfer Appliance is a physical device for large-scale data transfer to GCP. It is not suitable for live database migration with minimal downtime.

Cloud NAT

Why wrong: Cloud NAT provides outbound connectivity for private instances. It does not accelerate content delivery.

Cloud Armor

Why wrong: Cloud Armor is a web application firewall and DDoS protection service. It does not cache content.

Cloud DNS

Why wrong: Cloud DNS is a domain name system service. It does not cache web content.

Switch to BigQuery Editions with Autoscaling slots

Why wrong: Changing slot pricing models does not inherently improve query performance; it may increase cost without performance gain.

Use SELECT * only when necessary

Why wrong: Avoiding SELECT * is a best practice but does not address the performance issue for the described query.

Create a materialised view that pre-aggregates the data

Why wrong: Materialised views can improve performance but may increase storage costs. The question asks to optimise without increasing slot costs, but materialised views do not directly affect slot costs; they use storage. The question's focus is on performance optimisation without increasing slot costs, and materialised views are not strictly necessary if partitioning and clustering are applied.

Set a lifecycle rule to transition objects older than 90 days to Coldline storage class

Why wrong: Coldline is for data accessed less than once a quarter, not less than once a year. Archive is cheaper.

Rename objects to start with 'archive/' to reduce costs

Why wrong: Object naming does not affect storage costs. Cost is based on storage class and amount of data.

Set a lifecycle rule to transition objects older than 30 days to Nearline storage class

Why wrong: Nearline is for data accessed less than once a month, not less than once a year. It is more expensive than Archive.

Single-region configuration

Why wrong: Single-region does not provide failover to another region. An outage in the region would cause downtime.

Backup and restore in another region

Why wrong: Backup and restore would have an RPO of hours and RTO of hours, not meeting the 1-hour RPO and 15-minute RTO.

Dual-region with asynchronous replication

Why wrong: Cloud Spanner only supports synchronous replication. Asynchronous replication is not a feature.

Prepend a random number field: 'random_YYYYMMDD_userID_productID'

Why wrong: A random prefix distributes writes but makes range scans inefficient. For point reads, it may help distribution but is not optimal for time-range queries.

Keep the same order but use a shorter timestamp: 'YYMMDD_userID_productID'

Why wrong: Shortening the timestamp does not address the hotspot issue; recent dates still cluster together.

Reverse the timestamp: 'DDMMYYYY_userID_productID'

Why wrong: Reversing the timestamp does not change the fact that recent timestamps are similar; it may still cause hotspots if users read recent data.

Error rate

Why wrong: Error rate is related to availability but not directly. Availability is usually defined as the fraction of successful requests, not the rate of errors.

Request latency

Why wrong: Latency is a performance SLI, not availability. Availability measures whether the service is up and reachable.

CPU utilization

Why wrong: CPU utilization is a resource saturation metric, not a direct indicator of service availability.

Use Storage Transfer Service from another cloud

Why wrong: Storage Transfer Service is for cloud-to-cloud transfers, not for on-premises initial upload.

Use gsutil rsync over the network

Why wrong: At 100 Mbps, it would take about 190 days to transfer 200 TB, exceeding the 30-day window.

Use Cloud Data Fusion

Why wrong: Cloud Data Fusion is an ETL service, not designed for bulk file transfer to Cloud Storage.

On-demand pricing

Why wrong: On-demand pricing charges per query. For predictable queries, costs can be high compared to flat-rate for steady usage.

Use committed use discounts on slots

Why wrong: Committed use discounts apply to slot reservations but don't differentiate between scheduled and ad-hoc. They are an add-on to pricing models, not a pricing model themselves.

Flat-rate pricing (slot reservations) for all queries

Why wrong: Flat-rate pricing is cost-effective for large steady usage, but if ad-hoc queries are infrequent, unused slots are wasted.

Use Bigtable with replication

Why wrong: Bigtable provides eventual consistency across clusters. It does not offer strong consistency for multi-region deployments.

Use Firestore in Datastore mode

Why wrong: Datastore mode also provides eventual consistency across regions.

Use Firestore in multi-region (nam5 or eur3)

Why wrong: Firestore multi-region provides eventual consistency, not strong consistency.

Use cache invalidation after each update

Why wrong: Cache invalidation removes objects from Cloud CDN edge caches, but the default minimum TTL for Cloud CDN is 5 minutes, and the company updates content every 2 hours; invalidation alone cannot force the CDN to fetch fresh content from the bucket within 5 minutes if the TTL is set longer than that. This option is tempting because invalidation is designed to purge stale cached objects immediately after a change, and it would be correct if the requirement were to clear outdated data from edges without adjusting cache settings.

Use custom cache keys

Why wrong: Custom cache keys control cache grouping, not freshness.

Change cache mode to CACHE_ALL_STATIC

Why wrong: Cache mode affects what gets cached (e.g., static vs dynamic), not the TTL.

Use sole-tenant nodes to reduce risk of preemption

Why wrong: Sole-tenant nodes do not protect against preemption; they isolate VMs but preemptible VMs can still be preempted.

Switch to regular (non-preemptible) VMs

Why wrong: Regular VMs are more expensive and would increase cost, contradicting the requirement not to significantly increase cost.

Increase the number of preemptible VMs

Why wrong: Adding more VMs increases cost and does not directly improve reliability against preemption; it only increases parallelism.

Define SLOs and SLIs

Why wrong: SLOs and SLIs are used to set and measure reliability targets, not for incident review.

Create an error budget policy

Why wrong: Error budgets are used to balance reliability with feature velocity, not for incident postmortems.

Perform capacity planning

Why wrong: Capacity planning is about ensuring sufficient resources, not incident analysis.

Database Migration Service

Why wrong: DMS is for database migrations (e.g., MySQL to Cloud SQL), not VM disk migration.

Storage Transfer Service

Why wrong: Storage Transfer Service is for transferring data between cloud storage buckets, not VM disks.

Transfer Appliance

Why wrong: Transfer Appliance is for offline bulk data transfer, not live VM migration.