Dataproc for Spark/Hadoop in GCP - Step-by-Step Execution

Process Flow - Dataproc for Spark/Hadoop

User submits job

↓

Dataproc cluster receives job

↓

Job scheduled on cluster nodes

↓

Spark/Hadoop processes data

↓

Results stored in output location

↓

User retrieves results

This flow shows how a user submits a Spark or Hadoop job to a Dataproc cluster, which processes the data and stores the results.

Execution Sample

GCP

gcloud dataproc jobs submit spark --cluster=my-cluster --class=org.apache.spark.examples.SparkPi --region=us-central1 --jars=gs://dataproc-examples-2.0/jars/spark-examples_2.12-3.3.1.jar 1000

This command submits a Spark job to a Dataproc cluster to calculate Pi using 1000 samples.

Process Table

Step	Action	Input/Condition	Result/Output
1	Submit job	User runs gcloud command	Job sent to Dataproc cluster
2	Cluster receives job	Job arrives at cluster	Job queued for execution
3	Schedule job	Cluster resources available	Job assigned to nodes
4	Run Spark job	SparkPi class runs with 1000 samples	Pi calculated approximately
5	Store results	Job completes successfully	Output saved to storage
6	Retrieve results	User checks output location	User gets Pi result
7	Exit	Job finished	No more actions

💡 Job finishes after results are stored and retrieved

Status Tracker

Variable	Start	After Step 2	After Step 4	Final
Job Status	Not submitted	Queued	Running	Completed
Pi Value	N/A	N/A	3.14 approx	3.14 approx
Output Location	Empty	Empty	Empty	Contains result file

Key Moments - 3 Insights

Why does the job status change from 'Queued' to 'Running'?

Because the cluster assigns the job to nodes when resources become available, as shown in execution_table step 3 and 4.

Where are the results stored after the job completes?

Results are stored in the output location, typically cloud storage, as shown in execution_table step 5.

What happens if the cluster has no available resources?

The job stays in the queue until resources free up, delaying step 3 scheduling.

Visual Quiz - 3 Questions

Look at the execution_table, what is the job status after step 4?

A Completed

B Queued

C Running

D Not submitted

💡 Hint

Check the 'Job Status' variable in variable_tracker after step 4

At which step are the results stored in the output location?

A Step 5

B Step 4

C Step 3

D Step 6

💡 Hint

Look at execution_table row describing 'Store results'

If the user changes the sample size from 1000 to 10000, how would the execution_table change?

A Step 2 would be skipped

B Step 4 would take longer to run

C Step 5 would not store results

D Job would not be submitted

💡 Hint

Increasing samples affects processing time in step 4, see 'Run Spark job' action

Concept Snapshot

Dataproc runs Spark/Hadoop jobs on managed clusters.
User submits job via gcloud or API.
Cluster schedules and runs job on nodes.
Results saved to cloud storage.
Simple, scalable big data processing in the cloud.

Full Transcript

Dataproc lets users run Spark or Hadoop jobs easily on Google Cloud. The user submits a job, which the Dataproc cluster receives and queues. When resources are free, the cluster schedules the job on nodes. The Spark or Hadoop job processes data, for example calculating Pi with SparkPi. After processing, results are saved to cloud storage. The user can then retrieve the results. The job status changes from not submitted, to queued, to running, and finally completed. If resources are busy, the job waits in queue. Changing job parameters like sample size affects processing time. This flow simplifies big data processing by managing infrastructure automatically.

Practice

1. What is the main purpose of Google Cloud Dataproc when working with Spark and Hadoop?

easy

A. To manage and simplify running Spark and Hadoop clusters

B. To replace Spark and Hadoop with a new technology

C. To store data permanently like a database

D. To create virtual machines for web hosting

Solution

1. Step 1: Understand Dataproc's role

   Dataproc is a managed service that helps run Spark and Hadoop without manual setup.

2. Step 2: Compare options

   The other options describe replacing Spark and Hadoop, storing data permanently like a database, or creating VMs for web hosting, none of which is Dataproc's main purpose.

3. Final Answer:

   To manage and simplify running Spark and Hadoop clusters -> Option A

4. Quick Check:

   Dataproc manages Spark/Hadoop clusters = A [OK]

Hint: Dataproc manages clusters, not replaces or stores data [OK]

Common Mistakes:

• Thinking Dataproc replaces Spark/Hadoop
• Confusing Dataproc with storage services
• Assuming Dataproc creates VMs for web hosting

2. Which of the following is the correct gcloud command to create a Dataproc cluster named my-cluster in the us-central1 region?

easy

A. gcloud dataproc create cluster my-cluster --region us-central1

B. gcloud dataproc cluster create my-cluster --zone us-central1

C. gcloud dataproc clusters new my-cluster region=us-central1

D. gcloud dataproc clusters create my-cluster --region=us-central1

Solution

1. Step 1: Recall correct gcloud syntax

   The correct command uses 'clusters create' followed by the cluster name and '--region=' flag.

2. Step 2: Check each option

   'gcloud dataproc clusters create my-cluster --region=us-central1' matches the correct syntax. The other options have incorrect order, flags, or keywords.

3. Final Answer:

   gcloud dataproc clusters create my-cluster --region=us-central1 -> Option D

4. Quick Check:

   Correct gcloud cluster create syntax = B [OK]

Hint: Use 'clusters create' and '--region=' flag for cluster creation [OK]

Common Mistakes:

• Using 'create cluster' instead of 'clusters create'
• Missing equal sign in '--region='
• Confusing zone and region flags

3. Given the following command, what will be the result?

gcloud dataproc clusters create test-cluster --region=us-east1 --num-workers=2 --worker-machine-type=n1-standard-2

medium

A. Creates a cluster but in the default region ignoring '--region' flag

B. Creates a cluster named 'test-cluster' with 2 worker nodes of type n1-standard-2 in us-east1

C. Fails because '--num-workers' is not a valid flag

D. Creates a cluster with default worker count and machine type ignoring specified flags

Solution

1. Step 1: Analyze command flags

   The command specifies cluster name, region, number of workers, and worker machine type correctly.

2. Step 2: Understand Dataproc behavior

   Dataproc creates the cluster with given parameters if flags are valid and syntax is correct.

3. Final Answer:

   Creates a cluster named 'test-cluster' with 2 worker nodes of type n1-standard-2 in us-east1 -> Option B

4. Quick Check:

   Flags set cluster size and type correctly = D [OK]

Hint: Flags set cluster size and type; valid flags create cluster as specified [OK]

Common Mistakes:

• Assuming '--num-workers' is invalid
• Ignoring region flag effects
• Thinking default values override flags

4. You run this command to create a Dataproc cluster but get an error:

gcloud dataproc clusters create mycluster --num-worker=3 --region=us-west1

What is the likely cause of the error?

medium

A. Cluster name 'mycluster' is invalid because it has no hyphen

B. The region 'us-west1' is not supported by Dataproc

C. The flag '--num-worker' is incorrect; it should be '--num-workers'

D. Missing machine type flag causes the error

Solution

1. Step 1: Check flag correctness

   The correct flag for worker count is '--num-workers' (plural), not '--num-worker'.

2. Step 2: Validate other inputs

   Region 'us-west1' is valid, cluster name format is acceptable, and machine type is optional with defaults.

3. Final Answer:

   The flag '--num-worker' is incorrect; it should be '--num-workers' -> Option C

4. Quick Check:

   Correct flag spelling fixes error = C [OK]

Hint: Check plural form of flags like '--num-workers' [OK]

Common Mistakes:

• Using singular '--num-worker' instead of plural
• Assuming region is invalid without checking
• Thinking machine type is mandatory

5. You want to run a Spark job on a Dataproc cluster that automatically scales worker nodes between 2 and 5 based on workload. Which feature should you enable when creating the cluster?

hard

A. Enable autoscaling policy and attach it to the cluster

B. Set '--num-workers=5' and manually delete nodes when idle

C. Use '--max-workers=5' without autoscaling policy

D. Create multiple clusters with different sizes and switch between them

Solution

1. Step 1: Understand autoscaling in Dataproc

   Dataproc supports autoscaling policies to adjust worker nodes automatically within set limits.

2. Step 2: Evaluate options

   Enable autoscaling policy and attach it to the cluster correctly enables autoscaling with a policy. The other options do not provide automatic scaling: they either use fixed sizes, incomplete flags without policy, or require manual switching between clusters.

3. Final Answer:

   Enable autoscaling policy and attach it to the cluster -> Option A

4. Quick Check:

   Autoscaling policy enables dynamic scaling = A [OK]

Hint: Use autoscaling policies for automatic worker scaling [OK]

Common Mistakes:

• Setting fixed worker count without autoscaling
• Ignoring autoscaling policies
• Creating multiple clusters instead of scaling one