PCDE Practice Test 33 — 15 Questions

Question 1

A financial services company uses Cloud Spanner for a globally distributed application. They need to add a new column to an existing table that contains billions of rows. The change must not cause any downtime or performance degradation. How should the engineer implement this schema change?

Choose an answer to begin — your selection is scored in the full session.

Accepted Answer

Use ALTER TABLE ADD COLUMN; Spanner applies the change online without downtime.. Cloud Spanner supports online schema changes via the `ALTER TABLE ADD COLUMN` statement, which applies the change asynchronously without locking the table or causing downtime. Spanner uses a multi-phase approach to add columns, allowing reads and writes to continue uninterrupted while the schema change propagates across all nodes and regions.

Answer

Export the table, modify the schema in a temporary table, and import it back.

Answer

Use CREATE INDEX to add the column as an index.

Answer

Create a new table with the additional column, backfill data from the old table, then rename the tables.

Question 2

A DevOps engineer wants to create a dashboard that shows the number of 5xx errors per service over time. The errors are logged in Cloud Logging. What is the most efficient way to create this dashboard?

Accepted Answer

Create a log-based counter metric for 5xx errors, then add a chart in Cloud Monitoring using that metric.. The most efficient way to chart 5xx errors per service over time is to create a log-based counter metric in Cloud Logging that counts matching log entries, then chart that metric in Cloud Monitoring. Log-based metrics are computed by the logging service from the log stream, so no custom application code or data export is needed, and the metric is immediately available to Monitoring dashboards and alerting. This is the native, low-overhead integration path.

Answer

Write a custom application that sends the error count to Cloud Monitoring as a custom metric via API.

Answer

Export logs to BigQuery and use Data Studio to create a dashboard.

Answer

Use Cloud Logging's metrics explorer to create a chart and embed it in a dashboard.

Question 3

A company needs to store time-series sensor data with high write throughput (millions of writes per second) and low latency reads. Which database service should they choose?

Accepted Answer

Cloud Bigtable. Cloud Bigtable is a fully managed, scalable NoSQL database designed for large analytical and operational workloads, handling millions of writes per second with consistent low-latency reads. It uses a distributed, replicated SSTable storage engine and is optimized for time-series data, making it ideal for high-throughput sensor ingestion.

Answer

Cloud SQL

Answer

Firestore

Answer

Cloud Spanner

Question 4

A team wants to automate the response to a common incident: restarting a service when it becomes unhealthy. Which GCP service is best suited to trigger a Cloud Function based on a Cloud Monitoring alert?

Accepted Answer

Cloud Monitoring alert with notification channel to Pub/Sub. Cloud Monitoring alerts can publish to a Pub/Sub notification channel, and a Cloud Function can subscribe to that topic to execute remediation logic such as restarting a service. This decouples detection (Monitoring) from response (Function) and is the canonical event-driven remediation pattern on GCP. The alert policy defines the condition, and Pub/Sub provides the transport to the function.

Answer

Cloud Endpoints

Answer

Cloud Scheduler

Answer

Cloud Tasks

Question 5

Which TWO techniques can help avoid hot spotting in a Cloud Spanner table?

Accepted Answer

Add a hash of the primary key as the first part of the key. Adding a hash of the primary key as the first part of the key distributes writes evenly across Cloud Spanner's split boundaries. This prevents hot spotting, which occurs when monotonically increasing keys cause all new writes to land on the same tablet server. By hashing the key, writes are spread across multiple nodes, avoiding contention.

Answer

Use a monotonically increasing integer as the key

Answer

Use interleaved tables to distribute writes

Answer

Create a secondary index on a high-cardinality column

Question 6

A gaming company ingests player clickstream data in real time via Cloud Pub/Sub. They need to aggregate events per player session in BigQuery with exactly-once semantics. Which architecture minimizes latency and cost?

Accepted Answer

Use Cloud Dataflow with exactly-once processing to BigQuery. Cloud Dataflow with exactly-once processing is the correct choice because it provides a unified stream and batch processing model that guarantees exactly-once semantics when writing to BigQuery via the BigQuery I/O connector. This minimizes latency by processing events in micro-batches or streaming mode while avoiding duplicate data, and it is cost-effective as Dataflow auto-scales based on the Pub/Sub throughput.

Answer

Use Cloud Functions to write each message directly to BigQuery

Answer

Use Cloud Pub/Sub subscription to write to BigQuery directly

Answer

Use Cloud Dataproc to run Spark streaming jobs

Question 7

A team is using Skaffold for local development and CI/CD. They want to use Skaffold profiles to handle differences between local and production environments. In the CI pipeline (Cloud Build), they want Skaffold to build and deploy using the production profile. How should they invoke Skaffold in the cloudbuild.yaml?

Accepted Answer

skaffold run --profile production. The `skaffold run` command builds and deploys the application, and the `--profile` flag activates the specified profile. In a CI pipeline, using `skaffold run --profile production` ensures that the production profile's configurations (e.g., different image tags, deployment manifests) are applied during both build and deploy phases.

Answer

skaffold deploy --profile production

Answer

skaffold dev --profile production

Answer

skaffold build --profile production

Question 8

You are implementing a chaos engineering experiment on a GKE cluster using Chaos Mesh. You want to test the resilience of a microservice by injecting a 5-second delay into 50% of HTTP requests to a specific service. Which Chaos Mesh resource should you use?

Accepted Answer

HTTPChaos. HTTPChaos in Chaos Mesh is specifically designed to inject HTTP-level faults such as delays, aborts, and response modifications into requests matching defined criteria. To delay 50% of HTTP requests to a specific service, HTTPChaos with a delay action and a 50% probability selector is the correct resource.

Answer

NetworkChaos

Answer

Traffic Director fault injection

Answer

PodChaos

Question 9

A company is using DMS to migrate from MySQL to Cloud SQL. During the full dump phase, the migration is taking longer than expected. Which factor most likely affects the duration of the full dump?

Accepted Answer

Size of the source database. The duration of the full dump phase in a DMS migration is primarily determined by the size of the source database — the total volume of data that must be read from the source and written to the target. Larger databases require more time to transfer, regardless of engine version or table count. While other factors can influence performance, data size is the dominant factor in full load duration.

Answer

Source database engine version

Answer

Number of tables in the source

Answer

Replication lag during CDC

Question 10

A database administrator wants to set up monitoring and alerting for Cloud SQL instances. They need to be notified when CPU utilisation exceeds 80% for more than 5 minutes and when replication lag on a read replica exceeds 30 seconds. Which two metrics should they create alerting policies for? (Choose TWO.)

Accepted Answer

cloudsql.googleapis.com/database/replication/replica_lag. Option D, cloudsql.googleapis.com/database/cpu/utilization, is correct because it is the Cloud SQL-specific metric that reports CPU utilization as a percentage, which is exactly what the administrator needs to alert on when it exceeds 80% for more than 5 minutes. Option C, cloudsql.googleapis.com/database/replication/replica_lag, is correct because it directly measures the replication delay (in seconds) on a read replica, matching the requirement to alert when lag exceeds 30 seconds. Option A, cloudsql.googleapis.com/database/network/received_bytes_count, tracks network traffic volume rather than CPU or replication lag, so it does not address either alerting condition. Option B, cloudsql.googleapis.com/database/disk/bytes_used, measures disk space consumption, which is unrelated to the CPU and replication-lag thresholds. Option E, agent.googleapis.com/cpu_usage, is an Ops Agent metric for generic host CPU usage and is not the Cloud SQL database metric needed for instance-level CPU alerting.

Answer

cloudsql.googleapis.com/database/network/received_bytes_count

Answer

cloudsql.googleapis.com/database/disk/bytes_used

Answer

agent.googleapis.com/cpu_usage

Question 11

A DevOps engineer is designing a shared VPC topology for a multi-project environment. Which service project permission allows a project to use subnets from a host project?

Accepted Answer

compute.networkUser on the host project. In Google Cloud shared VPC, the service project must be granted the compute.networkUser role on the host project (or on specific subnets) to allow its users to use the shared subnets. This permission enables the service project's resources to be created in the host project's subnets. The role is typically granted to the service project's service account or users.

Answer

resourcemanager.projectIamAdmin on the service project

Answer

compute.networkAdmin on the host project

Answer

compute.instanceAdmin on the service project

Question 12

An online payment processing system uses Cloud SQL for MySQL with a 1 TB database. The system experiences high write throughput (~5000 writes/sec) and needs sub-10ms latency. The current instance has 8 vCPUs and 32 GB RAM. Which two metrics would indicate that the instance needs a larger tier? (Choose TWO.)

Accepted Answer

CPU utilization consistently above 80%.. Option B is correct because sustained CPU utilization above 80% on an 8-vCPU Cloud SQL instance indicates the vCPUs are saturated by the ~5000 writes/sec workload, so the instance cannot keep up and a larger tier with more vCPUs is needed to restore sub-10ms latency. Option C is correct because disk IOPS consistently hitting the instance's I/O limit means the storage throughput ceiling is the bottleneck; on Cloud SQL for MySQL, IOPS capacity scales with instance size and disk type, so moving to a larger tier raises the IOPS limit and relieves the write bottleneck. Option A is not a scaling signal because 50% memory usage leaves ample buffer pool headroom, well below the threshold where RAM pressure would force disk reads. Option D is not a problem indicator since 5 ms average query latency is already within the required sub-10ms target. Option E is not a bottleneck because 100 Mbps is trivial relative to the multi-Gbps network bandwidth available to an 8-vCPU Cloud SQL instance.

Answer

Memory usage is 50% of 32 GB.

Answer

Average query latency of 5 ms.

Answer

Network traffic is 100 Mbps.

Question 13

A company is designing a data warehouse for analytics. They need to store structured and semi-structured data, support SQL queries with sub-second performance on petabytes, and integrate with Data Studio. Which service should they consider?

Accepted Answer

BigQuery. BigQuery is a serverless, highly scalable data warehouse that supports SQL queries on petabyte-scale datasets with sub-second performance and integrates with Data Studio. Cloud Bigtable is a NoSQL database and does not support SQL. Cloud Spanner is designed for OLTP workloads, not analytics. Dataproc is a managed Hadoop/Spark service, not a data warehouse. Cloud SQL is for smaller relational datasets and cannot handle petabytes.

Answer

Cloud Bigtable

Answer

Cloud Spanner

Answer

Dataproc

Answer

Cloud SQL

Question 14

Your company runs a business intelligence (BI) dashboard on BigQuery that refreshes every hour. The dashboard queries are complex with multiple JOINs and aggregations. Recently, the queries started taking longer than 30 minutes, causing timeouts. You check the BigQuery monitoring and see that the slot utilization consistently reaches 100% during the dashboard refresh. The project uses a flat-rate pricing model with 1000 slots. Other team members run ad-hoc queries during the same period. What is the most effective action to improve the dashboard performance?

Accepted Answer

Create a separate reservation for the dashboard queries with a baseline of 500 slots and use a low priority job queue for ad-hoc queries.. Creating a separate reservation for the dashboard queries with a baseline of 500 slots ensures that the critical BI dashboard always has guaranteed compute capacity, preventing starvation by ad-hoc queries. Using a low-priority job queue for ad-hoc queries allows them to use any remaining idle slots without interfering with the dashboard's reserved slots. This directly addresses the 100% slot utilization and timeout issue without requiring query rewrites or schedule changes.

Answer

Rewrite the dashboard queries to use fewer joins and aggregations.

Answer

Increase the total number of slots to 2000 to provide more capacity for all queries.

Answer

Schedule the dashboard refresh to run at a different time when ad-hoc usage is low.

Question 15

A game development company wants to use Memorystore for Redis as a session store. They need to ensure that when memory is full, the least recently used keys are evicted first. Which eviction policy should they configure?

Accepted Answer

allkeys-lru. (allkeys-lru) is correct because the requirement is to evict the least recently used keys from the entire keyspace when memory is full. The 'allkeys-lru' policy applies the LRU algorithm to all keys, regardless of whether they have an expiry set, which matches the need to evict the least recently used keys first across the entire dataset.

Answer

volatile-lru

Answer

volatile-ttl

Answer

noeviction