Dataproc for Spark/Hadoop in GCP - Practice Problems & Coding Challenges

Challenge - 5 Problems

❓ Architecture

intermediate

Choosing the right cluster type for a batch Spark job

You need to run a batch Spark job on Google Cloud Dataproc. The job runs once a day and processes large data sets. Which cluster configuration is the best choice to optimize cost and performance?

A Create a permanent Dataproc cluster with autoscaling enabled and submit the job to it.

B Create a transient Dataproc cluster that is created before the job and deleted after the job finishes.

C Use a permanent Dataproc cluster without autoscaling and keep it running 24/7.

D Use a single-node Dataproc cluster with preemptible workers for cost savings.

💡 Hint

Think about cost efficiency for jobs that run once a day.

Transient clusters are created for the job and deleted after completion, saving costs for batch jobs that run periodically. Permanent clusters running 24/7 incur costs even when idle. Autoscaling helps but is more suited for variable workloads.

❓ Configuration

intermediate

Configuring autoscaling policy for Dataproc cluster

You want to configure an autoscaling policy for a Dataproc cluster to scale workers between 2 and 10 nodes based on YARN memory usage. Which configuration snippet correctly sets this policy?

A { "workerConfig": { "minInstances": 2, "maxInstances": 10 }, "basicAlgorithm": { "yarnConfig": { "scaleUpMemoryThresholdPercent": 50, "scaleDownMemoryThresholdPercent": 80 } } }

B { "workerConfig": { "minInstances": 10, "maxInstances": 2 }, "basicAlgorithm": { "yarnConfig": { "scaleUpMemoryThresholdPercent": 50, "scaleDownMemoryThresholdPercent": 80 } } }

C { "workerConfig": { "minInstances": 2, "maxInstances": 10 }, "basicAlgorithm": { "yarnConfig": { "scaleUpMemoryThresholdPercent": 80, "scaleDownMemoryThresholdPercent": 50 } } }

D { "workerConfig": { "minInstances": 2, "maxInstances": 10 }, "basicAlgorithm": { "yarnConfig": { "scaleUpMemoryThresholdPercent": 90, "scaleDownMemoryThresholdPercent": 90 } } }

💡 Hint

Scale up when usage is high, scale down when usage is low.

The minInstances and maxInstances define the scaling range. The scaleUpMemoryThresholdPercent should be higher than scaleDownMemoryThresholdPercent to avoid rapid scaling. Option C correctly sets min 2, max 10, scale up at 80%, scale down at 50%.

❓ security

advanced

Securing access to Dataproc clusters with least privilege

You want to ensure that only specific users can submit jobs to a Dataproc cluster, and no one else can access the cluster's VMs. Which approach best enforces this principle?

A Grant the users the 'roles/dataproc.jobUser' IAM role on the cluster and configure firewall rules to restrict SSH access to only admin IPs.

B Grant the users the 'roles/dataproc.editor' IAM role on the cluster and disable SSH access to the cluster VMs.

C Grant the users the 'roles/compute.instanceAdmin' role and allow SSH access to all cluster VMs.

D Grant the users the 'roles/dataproc.viewer' role and enable OS Login for all users.

💡 Hint

Least privilege means giving only the permissions needed to submit jobs, not full editing or VM control.

The 'roles/dataproc.jobUser' role allows users to submit jobs without full cluster editing rights. Restricting SSH access via firewall rules limits VM access to admins only, enforcing least privilege.

❓ service_behavior

advanced

Understanding job failure behavior on preemptible workers

You run a Spark job on a Dataproc cluster with preemptible worker nodes. During execution, some preemptible workers are terminated by Google Cloud. What is the expected behavior of the Spark job?

A The Spark job automatically retries tasks lost due to preemptible worker termination and completes successfully if enough workers remain.

B The Spark job immediately fails when any preemptible worker is terminated.

C The Spark job continues without retrying lost tasks, resulting in incomplete output.

D The Spark job pauses until the preemptible workers are replaced by new nodes.

💡 Hint

Think about how Spark handles worker node failures.

Spark is designed to handle worker failures by retrying lost tasks on remaining or new nodes. Preemptible workers can be terminated, but Spark retries tasks to complete the job if enough resources remain.

🧠 Conceptual

expert

Impact of network tags on Dataproc cluster firewall rules

You create a Dataproc cluster with network tags 'dataproc-cluster' and want to allow SSH access only from your office IP range. You create a firewall rule allowing TCP port 22 for source IPs in your office range and target tags 'dataproc-cluster'. However, you cannot SSH into the cluster VMs. What is the most likely cause?

A The firewall rule source IP range is incorrect and does not include your office IPs.

B The firewall rule priority is too low, so it is overridden by a deny-all rule.

C Dataproc clusters ignore network tags for firewall rules and require IP-based rules only.

D The cluster VMs do not have the 'dataproc-cluster' network tag assigned, so the firewall rule does not apply.

💡 Hint

Firewall rules with target tags only apply to VMs with matching network tags.

If the cluster VMs lack the specified network tag, the firewall rule targeting that tag does not apply, blocking SSH access. Assigning the correct network tag to the VMs fixes this.