GCP GCP-PCDE Free Practice Questions — Page 3

Correct Answer: A. roles/spanner.databaseAdmin for Person A roles/spanner.databaseReader for Person B roles/spanner.backupWriter for Application C

Using least privilege (PoLP) and separation of concerns (SoC) for Cloud Spanner IAM: DBA needs full database admin rights → roles/spanner.databaseAdmin Analyst only reads data for reports → roles/spanner.databaseReader Backup application only creates backups → roles/spanner.backupWriter A is correct because backupWriter allows creating backups without granting excessive admin rights — perfectly aligned with PoLP.B is wrong because backupAdmin grants both create AND delete backup permissions — too broad for an automated backup process.C is wrong because databaseReader for the backup application gives read-only data access but no backup creation rights at all.D is wrong because databaseUser for the analyst allows both reads and writes — too broad for a report-only role.

Correct Answer: D. Use granular instance sizing in Cloud Spanner and Autoscaler.

The issue is costs higher than expected but performance within SLA. Cloud Spanner charges per node or processing unit. The Google-recommended approach for cost optimization is granular instance sizing (use processing units instead of full nodes) combined with Autoscaler to scale dynamically.D is correct because processing units (minimum 100 PU vs 1000 PU per node) allow fine-grained cost control, and Autoscaler adjusts capacity based on load automatically.A is wrong because manual scaling is error-prone, operationally heavy, and not a Google best practice.B is wrong because interleaving improves performance by co-locating data, but doesn't reduce costs directly.C is wrong because the query optimizer improves execution efficiency but doesn't reduce instance costs when the instance is over-provisioned.

Correct Answer: A. Maintain a target of 23% CPU utilization by locating: cluster-a in zone us-central1-a cluster-b in zone europe-west1-d cluster-c in zone asia-east1-b

For multi-region Bigtable HA expansion to APAC, Google recommends maintaining CPU utilization below 35% for single-cluster or below 70% for replication, but for multi-cluster replication with HA, the target is 35% per cluster — however for a 3-cluster multi-region setup, Google recommends 23% CPU utilization per cluster (to handle traffic if one cluster fails, the remaining two can each absorb 50% more).A is correct because it places clusters in US, Europe, and Asia covering global distribution including APAC, with the correct 23% CPU target for a 3-cluster replication group ensuring HA.B is wrong because all three clusters are in the US — no APAC coverage despite the expansion requirement.C is wrong because 35% CPU target is for single-cluster setups, not 3+ cluster replication groups.D is wrong because two clusters in Asia and none in Europe doesn't provide balanced global coverage, and 35% is the wrong target for multi-cluster replication.

Correct Answer: A. Write your data into Bigtable and use Dataproc and the Apache Hbase libraries for analysis.

Requirements: 8 TB storage, millions of reads/writes per second, low latency, big data analysis tools.A is correct because Bigtable is purpose-built for high-throughput, low-latency workloads at petabyte scale, and integrates natively with Dataproc and Apache HBase libraries for big data analysis — a perfect fit.B is wrong because Cloud SQL doesn't scale to millions of requests per second and is not designed for big data analytics workloads.C is wrong because Memorystore (Redis) is an in-memory cache — it cannot durably store 8 TB and is not suitable for big data analysis.D is wrong because BigQuery is optimized for batch analytics, not low-latency real-time read/write at millions of requests per second.

Correct Answers: D. Promote high-cardinality attributes in multi-attribute primary keys.; E. Use bit-reverse sequential value as the primary key.

Cloud Spanner hotspots occur when rows are written sequentially to the same split boundary. Google recommends avoiding monotonically increasing keys. D is correct because high-cardinality attributes (many unique values) distribute writes evenly across splits, preventing hotspots. E is correct because bit-reverse sequential values take auto-increment IDs and reverse their bits, distributing them across the key space while preserving uniqueness — a Google-recommended pattern specifically for avoiding hotspots. A is wrong because auto-incrementing keys are the primary cause of hotspots in Cloud Spanner — all new writes go to the same split boundary. B is wrong because normalization improves data integrity but doesn't address key distribution hotspots. C is wrong because low-cardinality attributes (few unique values like status flags) concentrate writes, making hotspots worse.