PCDE Practice Exam 70 — 20 Questions

Question 1

You are using Cloud Profiler to identify performance bottlenecks. You notice a function that consumes significant CPU time. However, the profiler shows only a small percentage of samples in that function. What is the most likely cause?

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

Accepted Answer

The function is not hot enough to be sampled frequently due to the low overhead sampling rate of 0.5%.. Cloud Profiler uses statistical sampling and has a low overhead of about 0.5%. If a function appears with a small percentage, it may be that the sampling rate is too low to capture it accurately, or the function is not the main bottleneck. However, a common reason is that the profiler's sampling rate is low to keep overhead minimal.

Answer

You need to enable sampling for the specific function.

Answer

The profiler only profiles heap memory, not CPU.

Answer

The profiler is not configured to profile CPU usage.

Question 2

A company wants to implement security controls for Compute Engine VMs across their organization. Which THREE organization policies can enforce VM security? (Choose 3)

Accepted Answer

`compute.requireShieldedVm`. These three policies enforce VM security: external IP restriction, shielded VM requirement, and trusted image projects.

Answer

`storage.uniformBucketLevelAccess`

Answer

`iam.disableServiceAccountKeyCreation`

Question 3

You are monitoring a Cloud Spanner instance that is experiencing high CPU utilization (consistently above 70%). You want to identify the root cause. Which TWO metrics should you examine first? (Choose two.)

Accepted Answer

Average commit latency. Examining read and write throughput helps identify if the workload is pushing the instance. Analyzing commit latency and lock wait time reveals contention. Stale reads show replica lag but are not primary indicators of high CPU. Node count is configuration, not utilization.

Answer

Lock wait time

Answer

Stale read rate

Answer

Number of nodes

Question 4

A company is designing a schema for Cloud Bigtable to store user sessions. Access patterns: (1) read all sessions for a given user ID, and (2) read a specific session by session ID. The row key should support both patterns efficiently. Which row key design is MOST appropriate?

Accepted Answer

Use user_id#session_id as the row key. Using 'user_id#session_id' as the row key allows prefix scans on user_id to retrieve all sessions for a user, and exact lookups on the full key for a specific session. This is a common pattern for Bigtable.

Answer

Use session_id as the row key and store user_id as a column

Answer

Use a hash of user_id as row key prefix and session_id as suffix

Answer

Use user_id as the row key and store multiple session columns

Question 5

A social media company uses Cloud Spanner with a multi-region configuration. During a regional outage, automatic failover occurred, but some transactions that were in-flight at the time of failure were lost. What is the most likely reason for this data loss?

Accepted Answer

The multi-region configuration included read-only replicas in some regions, causing loss of recent writes. In a multi-region Spanner configuration, the leader region handles writes. If a transaction was committed but not yet replicated to other regions before the leader region failed, it could be lost if the new leader region does not have that data. However, Spanner's multi-region configurations are designed for RPO=0 (no data loss) when using read-write replicas. If read-only replicas are used in some regions, data loss can occur because those replicas do not participate in the voting. The most common cause of data loss in Spanner multi-region is a misconfiguration where not all regions have read-write replicas.

Answer

The database had schema changes that were not replicated to standby replicas

Answer

The transaction isolation level was set to read committed instead of serializable

Answer

The application did not retry failed transactions after the failover

Question 6

A company is migrating an application from Datastore to Firestore in Datastore mode. They need to ensure zero downtime during the migration. What is the recommended approach?

Accepted Answer

Upgrade the existing Datastore project to Firestore in Datastore mode using the Cloud Console.. Upgrading an existing Datastore project to Firestore in Datastore mode via the Cloud Console is a built-in, one-way migration that preserves the existing database name, indexes, and data without requiring any export/import or application code changes. This process is designed to be a live upgrade with no downtime, as Firestore in Datastore mode is fully backward-compatible with the Datastore API, allowing existing queries and transactions to continue uninterrupted during the transition.

Answer

Export all data from Datastore to Cloud Storage, then import into a new Firestore database.

Answer

Run both Datastore and Firestore in parallel, writing to both until migration is complete.

Answer

Create a new Firestore database and redirect traffic gradually.

Question 7

A DevOps engineer wants to visualize the 99th percentile latency of an HTTP endpoint over the past week using Cloud Monitoring dashboards. The metric is available as a distribution from Cloud Trace. Which chart type should they use to display this percentile over time?

Accepted Answer

Line chart. A line chart is best for showing a metric value (like 99th percentile latency) over time. Heatmaps are for distributions, scatter plots show correlation, and stacked bars show contributions of components. The percentile over time is a single value per time point, so a line chart is appropriate.

Answer

Heatmap

Answer

Stacked bar chart

Answer

Scatter chart

Question 8

A company is migrating a PostgreSQL database to Cloud SQL using DMS with continuous CDC. During cutover, the engineer checks DMS metrics and sees the replication lag is consistently 0 seconds. What is the next step to complete the migration?

Accepted Answer

Promote the destination Cloud SQL instance.. When the lag is 0, the source and target are in sync. The next step is to promote the destination (Cloud SQL) to make it the primary, which stops replication and allows writes.

Answer

Restart the migration job.

Answer

Increase the number of DMS worker nodes.

Answer

Delete the source database.

Question 9

Refer to the exhibit. A developer tries to connect to a Cloud SQL instance from a VM using the public IP. The connection fails with this error. What should the developer do to fix the connection?

Accepted Answer

Connect using the Cloud SQL Proxy.. The error indicates that the Cloud SQL instance does not have an authorized network allowing the VM's public IP. However, the correct fix is to use the Cloud SQL Proxy, which establishes a secure, authenticated tunnel to the instance without needing to authorize the VM's IP. The proxy handles IAM-based authentication and encryption, bypassing the public IP network authorization requirement entirely.

Answer

Enable SSL-only mode on the Cloud SQL instance.

Answer

Add the client IP to the authorized networks.

Answer

Change the instance to require SSL for private connections.

Question 10

You are designing a Cloud SQL for PostgreSQL instance for an OLTP application. The application typically handles 500 concurrent connections and the working set is 8 GB. You estimate a buffer pool of 4 GB. What minimum memory allocation should you choose?

Accepted Answer

15 GB. Cloud SQL PostgreSQL recommends max_connections = RAM_MB/16, but this is a soft limit. More importantly, memory must accommodate the working set and buffer pool. With 500 connections, 8 GB working set + 4 GB buffer pool = 12 GB, but PostgreSQL also needs overhead. A safe minimum is 15 GB, but the smallest Cloud SQL tier with >12 GB is 15 GB (e.g., db-custom-2-15360). However, among the options, 15 GB is the only viable choice. Note: max_connections formula suggests 15GB RAM gives ~960 connections, which covers 500.

Answer

12 GB

Answer

8 GB

Answer

26 GB

Question 11

A mobile app stores user profiles in Firestore. Users are spread globally. Which data model ensures low latency reads and writes?

Accepted Answer

Subcollections under a geographic region collection (e.g., /regions/{region}/users/{user}). It uses geographic region as the top-level collection key, which enables Firestore to co-locate user documents within the same region. This minimizes latency by ensuring that reads and writes for users in the same geographic area are served from a nearby Firestore instance, leveraging Firestore's automatic multi-region replication and strong consistency within a location.

Answer

A single collection containing all user documents

Answer

One document per user in a single collection with composite indexes

Answer

One collection per geographic region

Question 12

An organization uses Cloud SQL for MySQL and needs to perform disaster recovery testing by failing over to a cross-region read replica without impacting the primary instance. They want to validate the promotion process and measure the actual RTO. Which approach should be used?

Accepted Answer

Promote a cross-region read replica in an isolated project or non-production environment. The safest way to test DR without impacting the primary is to promote a cross-region read replica in a non-production environment or an isolated project. The promotion is manual and can be tested by simulating a failover scenario. Alternatively, you can clone the replica to a separate instance for testing, but the most direct test is promoting a replica that is not serving production traffic.

Answer

Create a clone of the primary instance in the target region and test failover

Answer

Use Cloud SQL’s switchover feature for HA instances

Answer

Run a gcloud command to promote the existing read replica in the production project

Question 13

You need to create a custom metric to monitor the number of user logins per minute. Which metric kind should you use?

Accepted Answer

CUMULATIVE. A CUMULATIVE metric measures a count over time that increases monotonically, such as total logins. GAUGE measures an instantaneous value, DELTA measures a value over an interval, and a counter is not a metric kind.

Answer

DELTA

Answer

GAUGE

Answer

Counter

Question 14

A company uses Cloud Build to build a container image with Kaniko. They want to speed up builds by caching the base image layers. Which configuration should they add to their cloudbuild.yaml?

Accepted Answer

Add '--cache=true' and '--cache-repo=us-central1-docker.pkg.dev/$PROJECT_ID/cache' to the Kaniko builder arguments.. Kaniko does not rely on the Docker daemon, so it cannot use Docker's native layer caching. Instead, Kaniko supports remote caching by pushing cached layers to a container registry. Adding `--cache=true` enables layer caching, and `--cache-repo` specifies the registry repository where cached base image layers are stored, allowing subsequent builds to reuse them and significantly reduce build time.

Answer

Add a 'docker pull' step before the Kaniko step to pre-pull the base image.

Answer

Use Docker's --layer-cache flag in the build step.

Answer

Configure Cloud Build to use a private pool with SSD persistent disks.

Question 15

A service uses Cloud SQL for MySQL. To test resilience, you want to inject latency into database queries. Which chaos engineering approach is most suitable on Google Cloud?

Accepted Answer

Use Traffic Director fault injection filter on the Envoy proxy sidecar. Traffic Director's fault injection filter on the Envoy proxy sidecar can inject latency into database queries by intercepting traffic at the service mesh layer. This approach allows controlled introduction of delays without modifying applications. Option A is incorrect because gcloud sql instances patch cannot add artificial delay. Option C is incorrect as Cloud SQL's built-in maintenance does not simulate latency in queries. Option D is incorrect because Cloud Functions throttling does not inject latency into database queries.

Answer

Use gcloud sql instances patch to add artificial delay

Answer

Use Cloud SQL's built-in maintenance to simulate latency

Answer

Deploy Cloud Functions to throttle the database connection

Question 16

An organization is designing a Cloud Spanner schema for a social media application. The application frequently queries for all posts by a specific user, and also updates the number of likes on a post. To ensure high performance and avoid hotspots, which TWO schema design principles should the team apply? (Choose two.)

Accepted Answer

Interleave the Post table under the User table using UserID as the first part of the primary key. Interleaving the Post table under the User table colocates posts with their user, making queries for a user's posts efficient by reducing distributed reads. Using a UUID for the post ID ensures writes are distributed across the cluster, avoiding hotspots from sequential keys like timestamps.

Answer

Use a secondary index on the Post table for UserID queries

Answer

Denormalize the like count into the User table to avoid joins

Answer

Use a monotonically increasing integer as the post ID to simplify indexing

Question 17

An SRE team is conducting a blameless postmortem after an outage. Which THREE elements should be included in the postmortem document? (Choose 3 answers)

Accepted Answer

Contributing factors (e.g., why the failure happened).. A good postmortem includes a timeline, contributing factors, and action items with owners. Blaming individuals is counterproductive, and root cause is often too simplistic; the focus should be on contributing factors.

Answer

Identification of the person responsible.

Answer

Single root cause.

Question 18

A BigQuery table is partitioned by ingestion time (pseudo column _PARTITIONTIME) and uses the default partition expiration of 90 days. A data engineer runs a DELETE statement to remove rows older than 100 days. Why does this query process more bytes than expected?

Accepted Answer

The DELETE statement must scan all partitions because it uses a condition that does not prune partitions.. The DELETE statement uses a condition that does not reference the partitioning column (_PARTITIONTIME) in a way that allows partition pruning. Since the table is partitioned by ingestion time, BigQuery must scan all partitions to evaluate the filter, even though the condition logically targets rows older than 100 days. This results in processing more bytes than expected, as the default partition expiration of 90 days does not reduce the scan scope when the WHERE clause does not leverage the partitioning column.

Answer

The table is not partitioned; it is clustered.

Answer

The DELETE statement does not use a WHERE clause on a clustering column.

Answer

The DELETE statement filters on a custom timestamp column instead of _PARTITIONTIME.

Question 19

Which TWO data type mappings are correct when converting Oracle data types to PostgreSQL?

Accepted Answer

Oracle CLOB → PostgreSQL TEXT. Oracle's CLOB stores large variable-length character data, and PostgreSQL's TEXT type is the direct equivalent, supporting up to 1 GB of character data without length limitations. Option D is correct because Oracle's NUMBER(10) without scale defaults to a whole number that fits into PostgreSQL's INTEGER (32-bit range). Options B, C, and E are incorrect: VARCHAR2(100) maps to PostgreSQL VARCHAR(100) or TEXT, not CHAR(100) (fixed-length); NUMBER(10,2) requires DECIMAL or NUMERIC to preserve precision; Oracle's DATE includes time components and maps to TIMESTAMP, not DATE.

Answer

Oracle VARCHAR2(100) → PostgreSQL CHAR(100)

Answer

Oracle NUMBER(10,2) → PostgreSQL INTEGER

Answer

Oracle DATE → PostgreSQL DATE

Question 20

Your Cloud Run service processes requests from an external API that sends a burst of up to 100 requests per second. You want to maximize throughput while minimizing instances. The service is CPU-bound. What configuration should you use?

Accepted Answer

Set concurrency to 1 to give each request full CPU, and enable CPU always-on.. For CPU-bound services, the default concurrency of 80 may be too high, causing resource contention and slowing down each request. Reducing concurrency to a lower value, such as 1, dedicates the entire CPU to each request, potentially improving throughput per instance. CPU always-on is needed for background tasks but not necessarily for request handling. Gen2 execution environment can help with higher memory but not CPU-bound throughput directly.

Answer

Use the Gen2 execution environment with higher memory.

Answer

Set concurrency to 1000 to handle burst efficiently.

Answer

Set concurrency to 80 (default) and CPU always-on.