PCDE Practice Exam 51 — 20 Questions

Question 1

Your team runs a Cloud SQL for PostgreSQL instance that serves an internal expense reporting application. Over the past week, users have complained about slow response times during the morning submission window. You notice that CPU utilization on the primary instance spikes to 95% during that window, and you want to determine which SQL statements are contributing most to that CPU load. What should you do first?

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

Accepted Answer

Enable the pg_stat_statements extension and query it to identify statements with the highest total execution time.. The goal is to pinpoint which SQL statements consume the most CPU during the morning window. pg_stat_statements aggregates execution time per statement, directly exposing the top consumers. Raising vCPUs or enabling slow query logging does not provide per-statement aggregation, and instance-level metrics lack statement granularity. Querying pg_stat_statements is the most direct and lowest-risk diagnostic step.

Answer

Use Cloud Monitoring to view the CPU utilization metric for the instance and correlate it with the number of active connections.

Answer

Increase the number of vCPUs on the instance to reduce CPU utilization during the spike.

Answer

Enable slow query logging with a threshold of 100 ms and wait for the next morning window.

Question 2

A Cloud SQL for PostgreSQL 14 instance is used by a SaaS application and shows a steady increase in active connections over several weeks, reaching the max_connections limit during business hours. Users see 'FATAL: sorry, too many clients already' errors. The application uses a connection pool per pod, and there are 40 pods. You want to eliminate the connection errors without reducing application availability. What should you do?

Accepted Answer

Deploy a connection pooler such as PgBouncer in transaction pooling mode between the application and the instance, and point the pods at it.. Connection exhaustion from many application pods is best solved by inserting a transaction-mode pooler that multiplexes client sessions onto a small backend pool. This keeps PostgreSQL below max_connections and preserves throughput, whereas raising the limit risks resource exhaustion and shrinking pool sizes degrades concurrency. Monitoring extensions do not reduce connection demand.

Answer

Enable the pg_stat_statements extension and use it to identify and terminate long-running queries that are holding connections.

Answer

Increase max_connections on the instance to a value that accommodates all pods plus administrative headroom.

Answer

Lower the application's connection pool maximum size per pod so the total stays under max_connections.

Question 3

A company has a Spanner instance with 5 nodes serving a global application. They receive alerts that write latency has increased significantly during business hours in the Asia-Pacific region. The team confirms that no application changes have been made. What is the most likely cause and recommended action?

Accepted Answer

Writes are hitting a hot spot due to monotonically increasing keys; consider using a hash prefix or bit-reversed key. Monotonically increasing keys (e.g., timestamps or auto-increment IDs) cause all new writes to target the same tablet leader in Spanner, creating a hot spot. This increases write latency because the single node becomes a bottleneck, especially during peak business hours in the Asia-Pacific region. Using a hash prefix or bit-reversed key distributes writes evenly across nodes, resolving the contention.

Answer

CPU utilization is above 70%; enable Spanner fine-grained access control

Answer

Set up interleaved indexes to speed up writes

Answer

The instance is under-provisioned; increase the number of nodes

Question 4

A team is designing a Cloud Bigtable schema for storing time-series data from IoT sensors. Each row represents a sensor reading, with a row key composed of sensor ID and timestamp. They need to efficiently retrieve the latest reading for a specific sensor. What row key design should they use?

Accepted Answer

Use sensor ID as a prefix followed by an inverted timestamp (e.g., sensorID#Long.MAX_VALUE - timestamp).. In Cloud Bigtable, row keys are sorted lexicographically. Using sensor ID as a prefix groups all readings for a sensor, and an inverted timestamp ensures the latest reading appears first. This allows a single row lookup or a short scan to retrieve the latest reading, optimizing the required access pattern. It also distributes writes across sensors, avoiding hotspots.

Answer

Use timestamp as a prefix followed by sensor ID (e.g., timestamp#sensorID).

Answer

Use a random UUID as the row key to distribute writes evenly.

Answer

Use sensor ID as a prefix followed by a timestamp in ascending order.

Question 5

A retail company is designing a new inventory management system on Cloud Spanner. They need to ensure high write throughput for order processing. Which two schema design practices help avoid write hotspots? (Choose TWO.)

Accepted Answer

Avoid using a monotonically increasing primary key.. Option B is correct because Cloud Spanner splits data by primary key ranges, and a monotonically increasing key (such as a timestamp or auto-incrementing integer) causes all new writes to land on the last split, creating a hotspot; avoiding such keys spreads inserts across splits. Option E is correct because adding a hash prefix to the primary key randomizes key distribution, so consecutive inserts are scattered across many splits and write throughput scales horizontally. Option A is not correct here because secondary indexes add write amplification and index hotspots rather than preventing hotspots on the base table. Option C is not correct because storing everything in one non-interleaved table does nothing to distribute key ranges and can worsen contention. Option D is not correct because foreign keys enforce referential integrity and do not affect how rows are distributed across splits.

Answer

Create secondary indexes on frequently queried columns.

Answer

Store all data in a single table with no interleaving.

Answer

Use foreign keys to enforce referential integrity.

Question 6

You are managing a Cloud BigQuery dataset that contains a large fact table partitioned by ingestion time. Queries that filter on a timestamp column other than the partitioning column are performing full table scans and consuming many slots. You want to reduce the amount of data scanned and improve performance. What should you do?

Accepted Answer

Change the table partitioning to use the timestamp column as the partitioning key.. BigQuery prunes partitions based on the partitioning column. If queries filter on a different timestamp column, the partitioning column is not used, so no pruning occurs, resulting in a full table scan. Repartitioning the table by the timestamp column used in filters ensures that BigQuery can eliminate irrelevant partitions, reducing data scanned and improving performance.

Answer

Create a clustered table on the timestamp column.

Answer

Enable the use of cached results for the queries.

Answer

Increase the number of slots in the project to handle the full table scans.

Question 7

A company wants to implement a CI/CD pipeline for a multi-service application where each service is built from a separate repository. They need to run unit tests, build container images, and deploy to both Cloud Run and GKE. Which two Google Cloud services should be combined to achieve this?

Accepted Answer

Cloud Build. Cloud Build (C) is correct because it is Google Cloud's managed CI/CD service that can compile code, run unit tests, build container images, and push them to Artifact Registry, supporting multi-repository pipelines via triggers and build configs. Cloud Deploy (E) is correct because it is the managed continuous delivery service designed to take built container images and progressively deploy them to targets such as Cloud Run and GKE using delivery pipelines, releases, and rollouts. Together they cover the required build/test and deploy stages across both runtimes. Config Sync (A) is a GitOps tool for continuously reconciling Kubernetes cluster state, not for building or testing code. Artifact Registry (B) only stores and manages container images and packages; it is a dependency of the pipeline but does not itself run tests or orchestrate deployments. Cloud Source Repositories (D) is a private Git repository service for hosting source code, not a CI/CD build or deployment engine.

Answer

Config Sync

Answer

Artifact Registry

Answer

Cloud Source Repositories

Question 8

Which Cloud Monitoring metric indicates the number of queries waiting for locks in Cloud SQL?

Accepted Answer

Lock waits. The 'Lock waits' metric in Cloud SQL (for MySQL, PostgreSQL, or SQL Server) directly tracks the number of queries that are blocked because they are waiting for a lock held by another transaction. This is the correct indicator of query contention, as it measures the count of statements currently in a lock-wait state, not the total queries or connections.

Answer

Active connections

Answer

CPU utilization

Answer

Queries

Question 9

A BI manager needs to restrict access to sensitive sales data so that salespeople can only see their own region's data. Which BigQuery feature should be used to implement row-level security without duplicating tables?

Accepted Answer

Use BigQuery row-level access policies. BigQuery row-level access policies are a native feature that allows restricting access to specific rows in a table without duplicating data. They provide a direct, scalable, and efficient way to implement row-level security based on user attributes like region. While authorized views with SESSION_USER() can also achieve similar functionality, they are not the native feature and require additional management overhead. Therefore, Option B is the correct answer.

Answer

Use column-level security to hide sensitive columns

Answer

Create an authorized view that uses SESSION_USER() in a WHERE clause to filter rows

Answer

Create separate IAM roles for each region

Question 10

A startup is building a mobile app using Firestore in Native mode. The app stores user profiles in a collection 'users'. Each user document contains fields: userId (string), displayName (string), email (string), and lastLogin (timestamp). The app needs to display a list of users sorted by lastLogin in descending order, 20 at a time. What should you do to ensure the queries are efficient and scalable?

Accepted Answer

Use a query with orderBy('lastLogin', 'desc') and limit(20), relying on automatic single-field indexes.. Firestore automatically indexes all fields, so a query ordering by lastLogin descending and limiting results will use the single-field index on lastLogin. This approach is efficient, scalable, and requires no additional indexes or data duplication. Pagination can be implemented using the last document snapshot as a cursor for the next page.

Answer

Store a separate collection 'users_by_lastLogin' where each document ID is the lastLogin timestamp.

Answer

Create a composite index on lastLogin and userId.

Answer

Add a field 'lastLoginDesc' that stores the negative of the timestamp to allow ascending order.

Question 11

Which THREE are components of an effective incident management process? (Choose 3.)

Accepted Answer

Blameless postmortem with action items. Option B is correct because a blameless postmortem with action items is a core part of incident management: it focuses on systemic causes rather than individual fault, and it produces tracked remediation work that reduces recurrence. Option D is correct because an on-call rotation with defined escalation paths ensures incidents are acknowledged promptly and escalated to the right expertise when the primary responder cannot resolve them within expected timeframes. Option E is correct because the incident commander role provides a single accountable coordinator who directs response, delegates tasks, and maintains communication, preventing confusion during high-pressure incidents. Option A is not a standard component: automatic rollback of all changes is a risky, context-dependent remediation tactic, not a required element of the incident management process. Option C is not a component of incident management: monthly performance bonuses for on-call are a compensation or incentive policy and do not contribute to detecting, responding to, or learning from incidents.

Answer

Automatic rollback of all changes during incidents

Answer

Monthly performance bonuses for on-call

Question 12

A DevOps team is designing a landing zone on GCP. They want to centralize networking, logging, and security. Which TWO projects should they create? (Choose 2)

Accepted Answer

A central Logging project. Option B is correct because a central Logging project is the standard GCP landing zone pattern for aggregating logs from all projects into a single destination, where you configure log sinks, log buckets, and Cloud Logging retention/export policies so that audit and platform logs are centrally owned and not siloed per workload. Option C is correct because a Shared VPC host project centralizes networking: the host project owns the VPC network, subnets, firewall rules, and Cloud NAT/Cloud Router, while service projects attach as service projects and consume those shared subnets, which is exactly how a landing zone centralizes networking. The unmarked options do not belong: IAM roles are not resources that live in their own project (they are defined globally and bound to resources), and creating a project per environment or per application is a workload-placement decision that fragments rather than centralizes networking, logging, and security.

Answer

A project for IAM roles

Answer

A project for each environment

Answer

A separate project per application

Question 13

You have a Cloud Spanner database that needs to be migrated from one region to another. You want to ensure no data loss and minimal downtime. Which approach should you use?

Accepted Answer

Create a backup of the database and restore it to a new instance in the target region.. Cloud Spanner does not support in-place region changes or read replicas in different regions. The only supported method to migrate a Cloud Spanner database between regions with no data loss and minimal downtime is to create a backup of the database and restore it to a new instance in the target region. This approach ensures a consistent snapshot of the data and allows you to plan the cutover window to minimize downtime.

Answer

Use gcloud spanner instances move to change the region of the existing instance.

Answer

Create a read replica in the target region and promote it after replication catches up.

Answer

Use gcloud command 'gcloud spanner databases export' to export the database to CSV files, then import into the new instance.

Question 14

An engineer needs to reduce toil. Which of the following tasks is considered toil according to SRE principles?

Accepted Answer

Manually rotating service account keys every week. Toil is defined in SRE as manual, repetitive, automatable, tactical work that scales linearly with service growth and provides no enduring value. Manually rotating service account keys every week fits this definition precisely: it is repetitive, automatable via secrets managers or key rotation policies, and grows with the number of accounts. The other options involve engineering judgment, learning, or process improvement, which are not toil.

Answer

Writing a postmortem after an incident

Answer

Designing a new microservice architecture

Answer

Reviewing code changes for a new feature

Question 15

During a blameless postmortem, the team uses the '5 Whys' technique to identify root causes. What is the primary purpose of this technique?

Accepted Answer

To iteratively ask 'why' to uncover the root cause. The '5 Whys' technique is a root cause analysis method where you iteratively ask 'why' (typically five times) to drill past surface symptoms and uncover the underlying systemic cause of an incident. Its primary purpose is to move from the immediate cause to the root cause so that fixes prevent recurrence.

Answer

To estimate the cost of the incident

Answer

To identify the immediate superficial cause of the incident

Answer

To determine which team member made an error

Question 16

A batch job on GKE needs to be resilient to node failures. The job creates several pods that run for a few minutes each. The team wants to ensure that during a voluntary node disruption (e.g., node upgrade), only a limited number of pods are affected. Which resource should they configure?

Accepted Answer

Pod Disruption Budget. PodDisruptionBudget (PDB) specifies the minimum number of pods that must remain available during voluntary disruptions, protecting the job.

Answer

Vertical Pod Autoscaler

Answer

Horizontal Pod Autoscaler

Answer

Cluster Autoscaler

Question 17

Your Cloud SQL for PostgreSQL database is experiencing slow query performance. You want to identify and optimize the slowest queries. Which TWO actions should you take? (Choose two)

Accepted Answer

Enable the slow query log by setting the log_min_duration_statement flag.. Option C is correct because setting the log_min_duration_statement flag on Cloud SQL for PostgreSQL enables the slow query log, which records every statement whose execution time exceeds the specified threshold (in milliseconds), giving you the concrete list of slowest queries to target. Option E is correct because EXPLAIN ANALYZE actually executes the query and returns the real execution plan with per-node timing, row counts, and loop counts, which is the standard way to pinpoint bottlenecks such as sequential scans, bad join strategies, or missing indexes. Options A and B address scalability and connection overhead rather than identifying or optimizing the slowest queries: PgBouncer connection pooling reduces connection churn, and read replicas offload read traffic, but neither reveals which queries are slow or why. Option D, increasing instance memory to the next tier, is a blunt resource upgrade that may mask symptoms without diagnosing the offending queries, so it does not satisfy the goal of identifying and optimizing them.

Answer

Set up connection pooling using Cloud SQL Auth Proxy and PgBouncer.

Answer

Create read replicas to offload read traffic.

Answer

Increase the instance memory to the next tier.

Question 18

A DevOps team needs to set up alerting for a critical application that runs on Compute Engine. They want to be notified if the application process crashes (i.e., stops sending heartbeats) for more than 5 minutes. The application emits a custom metric 'app_heartbeat' as a GAUGE with value 1 every 60 seconds. Which TWO configurations should they use? (Choose 2)

Accepted Answer

Duration: 5 minutes. Option B (Duration: 5 minutes) is correct because the alerting policy's condition must remain true for the required 5-minute window before firing, matching the requirement to be notified only if the process stops heartbeating for more than 5 minutes. Option E (Metric absent condition: for 'app_heartbeat' metric) is correct because a crashed process stops emitting the custom metric entirely, so the appropriate Cloud Monitoring condition type is a metric-absence condition that triggers when no data points arrive for the specified duration. Options A, C, and D do not belong: an alignment period of 60 seconds (A) and a MEAN reducer (D) are relevant only to threshold conditions on received data, and a threshold condition of < 1 (C) would not fire when the metric is simply missing, since absent data is not evaluated as a value below the threshold.

Answer

Alignment period: 60 seconds

Answer

Metric threshold condition: threshold < 1

Answer

Reducer: MEAN

Question 19

A company wants to migrate their on-premises SQL Server database to Cloud SQL for PostgreSQL using Database Migration Service. They need to minimize downtime. The source database is 2 TB and the network link has 1 Gbps bandwidth. What should they do first?

Accepted Answer

Create a continuous migration job with DMS using a VPC peering connection.. For a 2 TB database migration with minimal downtime, a continuous migration job using Database Migration Service (DMS) with VPC peering is the correct first step. DMS continuous migration performs an initial full dump followed by ongoing change data capture (CDC) replication, keeping the target in sync with the source until cutover. This minimizes downtime because the bulk of the data is transferred while the source remains operational, and only the final cutover requires a brief application pause.

Answer

Use pg_dump to export the database and then import into Cloud SQL.

Answer

Create a one-time migration job to copy the database during a maintenance window.

Answer

Use mysqldump to export and import the database.

Question 20

Which two practices are characteristic of a blameless postmortem? (Choose TWO.)

Accepted Answer

Creating action items with owners and deadlines. Option A is correct because a blameless postmortem must produce concrete follow-up work: action items with assigned owners and due dates ensure the contributing factors identified are actually remediated and tracked to closure. Option D is correct because the core of blameless analysis is examining systemic and process contributing factors — such as missing safeguards, unclear runbooks, or inadequate monitoring — rather than individual fault, which is what allows real prevention. Option B is wrong because punitive measures are the opposite of a blameless culture; punishment suppresses reporting and hides the systemic causes. Option C is wrong because naming the employee who erred individualizes the failure and contradicts the blameless principle. Option E is wrong because postmortems are typically shared broadly (often sanitized and published internally) so the whole organization can learn; confidentiality within the incident response team defeats that learning purpose.

Answer

Assigning punitive measures to prevent recurrence

Answer

Identifying the employee who made the error

Answer

Keeping the postmortem confidential within the incident response team