YARN scheduling policies in Hadoop - Step-by-Step Execution

Concept Flow - YARN scheduling policies

Job arrives in YARN

↓

Scheduler selects policy

↓

FIFO: Jobs run in order

↓

Capacity: Resources split by queues

↓

Fair: Resources shared fairly

↓

Allocate resources to containers

↓

Run tasks

↓

Job completes or waits

YARN receives jobs and chooses a scheduling policy (FIFO, Capacity, or Fair) to allocate resources and run tasks.

Execution Sample

Hadoop

jobs = [Job1, Job2, Job3]
scheduler = 'FIFO'
for job in jobs:
    allocate_resources(job)
    run(job)

This code runs jobs one by one in the order they arrive using FIFO scheduling.

Execution Table

Step	Job	Scheduler Policy	Action	Resources Allocated	Status
1	Job1	FIFO	Allocate resources	4 containers	Running
2	Job1	FIFO	Run tasks	4 containers	Running
3	Job1	FIFO	Complete	0 containers	Finished
4	Job2	FIFO	Allocate resources	3 containers	Running
5	Job2	FIFO	Run tasks	3 containers	Running
6	Job2	FIFO	Complete	0 containers	Finished
7	Job3	FIFO	Allocate resources	2 containers	Running
8	Job3	FIFO	Run tasks	2 containers	Running
9	Job3	FIFO	Complete	0 containers	Finished

💡 All jobs completed in FIFO order, no more jobs to schedule.

Variable Tracker

Variable	Start	After Step 1	After Step 4	After Step 7	Final
Job1 Status	Waiting	Running	Finished	Finished	Finished
Job2 Status	Waiting	Waiting	Running	Finished	Finished
Job3 Status	Waiting	Waiting	Waiting	Running	Finished
Resources Allocated	0	4	3	2	0

Key Moments - 2 Insights

Why does Job2 wait until Job1 finishes in FIFO scheduling?

Because FIFO runs jobs strictly in arrival order, Job2 cannot start until Job1 completes, as shown in steps 1-6 in the execution_table.

How does resource allocation differ between Capacity and Fair schedulers?

Capacity divides resources by fixed queues, while Fair shares resources dynamically to balance usage. This differs from FIFO's one-job-at-a-time approach.

Visual Quiz - 3 Questions

Look at the execution_table, what is the status of Job2 at Step 3?

A Waiting

B Running

C Finished

D Not started

💡 Hint

Check the 'Status' column for Job2 at Step 3 in the execution_table.

At which step does Job3 start running in FIFO scheduling?

A Step 4

B Step 7

C Step 6

D Step 9

💡 Hint

Look at the 'Action' and 'Status' columns for Job3 in the execution_table.

If we switch to Fair scheduler, how would resource allocation change?

A Jobs run one by one with full resources

B Only the first job gets resources

C Resources are shared among jobs simultaneously

D Jobs run in reverse order

💡 Hint

Fair scheduler shares resources fairly among running jobs, unlike FIFO.

Concept Snapshot

YARN Scheduling Policies:
- FIFO: Jobs run one after another in arrival order.
- Capacity: Resources split by queues with fixed capacity.
- Fair: Resources shared dynamically for fairness.
- Scheduler allocates containers to run tasks.
- Choice affects job wait and resource use.

Full Transcript

YARN scheduling policies decide how jobs get resources to run. FIFO runs jobs one by one in order. Capacity scheduler splits resources by queues with set limits. Fair scheduler shares resources evenly among jobs. This example shows FIFO running three jobs in order, allocating containers, running tasks, and finishing each job before starting the next. Variables track job status and resource use. Key points include why jobs wait in FIFO and how resource sharing differs in other policies. Quizzes check understanding of job status at steps and scheduler behavior.

Practice

1. Which YARN scheduling policy assigns resources to jobs in the order they arrive without considering job size or priority?

easy

A. Capacity Scheduler

B. Fair Scheduler

C. FIFO Scheduler

D. Priority Scheduler

Solution

1. Step 1: Understand FIFO Scheduler behavior

   FIFO (First In First Out) scheduler assigns resources to jobs strictly by arrival time, ignoring job size or priority.

2. Step 2: Compare with other schedulers

   Capacity Scheduler and Fair Scheduler consider resource sharing and fairness, while Priority Scheduler uses job priority.

3. Final Answer:

   FIFO Scheduler -> Option C

4. Quick Check:

   FIFO = order by arrival [OK]

Hint: FIFO means first come, first served order [OK]

Common Mistakes:

• Confusing FIFO with Fair Scheduler
• Thinking Capacity Scheduler ignores job order
• Assuming Priority Scheduler is default

2. Which of the following is the correct syntax to configure the Capacity Scheduler in YARN's capacity-scheduler.xml file?

easy

A. yarn.scheduler.capacity.root.queuesdefault,production

B. yarn.scheduler.fifo.root.queuesdefault,production

C. yarn.scheduler.fair.root.queuesdefault,production

D. yarn.scheduler.priority.root.queuesdefault,production

Solution

1. Step 1: Identify correct property prefix for Capacity Scheduler

   The Capacity Scheduler uses properties starting with yarn.scheduler.capacity.

2. Step 2: Verify queue configuration syntax

   The property yarn.scheduler.capacity.root.queues defines queues like default and production.

3. Final Answer:

   <property><name>yarn.scheduler.capacity.root.queues</name><value>default,production</value></property> -> Option A

4. Quick Check:

   Capacity Scheduler uses yarn.scheduler.capacity prefix [OK]

Hint: Capacity Scheduler uses 'yarn.scheduler.capacity' prefix [OK]

Common Mistakes:

• Using fifo or fair prefix for Capacity Scheduler
• Misspelling property names
• Confusing queue names with scheduler types

3. Given the following YARN Fair Scheduler configuration snippet:
<pre>queueA: minResources=2GB, maxResources=10GB queueB: minResources=4GB, maxResources=8GB Currently, queueA is using 3GB and queueB is using 5GB.
Which queue can get more resources if a new job requests 3GB?

medium

A. queueA only

B. Both queueA and queueB

C. queueB only

D. Neither queueA nor queueB

Solution

1. Step 1: Check current usage vs maxResources

   queueA uses 3GB, max is 10GB, so it can get up to 7GB more. queueB uses 5GB, max is 8GB, so it can get 3GB more.

2. Step 2: Assess if requested 3GB fits

   queueA can allocate 3GB (3+3=6GB <= 10GB). queueB can allocate 3GB (5+3=8GB = max), so it can also get 3GB.

3. Step 3: Consider Fair Scheduler policy

   Fair Scheduler tries to balance usage. Since queueA is below queueB usage, it will prioritize queueA for extra resources, but queueB can also allocate the requested resources up to its max.

4. Final Answer:

   Both queueA and queueB -> Option B

5. Quick Check:

   Fair Scheduler favors less used queue but both can allocate requested resources [OK]

Hint: Fair Scheduler favors queues with less current usage [OK]

Common Mistakes:

• Ignoring maxResources limits
• Assuming both queues get resources equally
• Not considering current usage in decision

4. You have configured the Capacity Scheduler but notice that some queues are starved and not getting resources. What is the most likely cause?

medium

A. Queue capacities are not properly set, causing imbalance

B. FIFO Scheduler is enabled instead of Capacity Scheduler

C. The yarn.scheduler.capacity.root.queues property is missing

D. Fair Scheduler is overriding Capacity Scheduler settings

Solution

1. Step 1: Understand Capacity Scheduler behavior

   Capacity Scheduler allocates resources based on configured queue capacities.

2. Step 2: Identify cause of starvation

   If queue capacities are not set correctly, some queues may get more resources, starving others.

3. Step 3: Rule out other options

   Missing property would cause scheduler failure, not starvation. FIFO and Fair Scheduler are different policies, not overriding Capacity Scheduler if configured correctly.

4. Final Answer:

   Queue capacities are not properly set, causing imbalance -> Option A

5. Quick Check:

   Capacity settings control resource distribution [OK]

Hint: Check queue capacity percentages to avoid starvation [OK]

Common Mistakes:

• Assuming missing property causes starvation
• Confusing scheduler types
• Ignoring capacity configuration importance

5. You want to run a multi-tenant Hadoop cluster where each department gets guaranteed minimum resources but can use more if available. Which YARN scheduling policy and configuration best fits this need?

hard

A. Priority Scheduler with fixed job priorities

B. FIFO Scheduler with strict job ordering

C. Fair Scheduler with no minimum resource guarantees

D. Capacity Scheduler with minimum and maximum queue capacities set

Solution

1. Step 1: Understand multi-tenant resource needs

   Departments need guaranteed minimum resources but flexible usage if available.

2. Step 2: Match policy features

   Capacity Scheduler supports minimum and maximum queue capacities, ensuring guarantees and flexible sharing.

3. Step 3: Rule out other policies

   FIFO does not guarantee resources per tenant. Fair Scheduler can share fairly but may lack strict minimum guarantees. Priority Scheduler focuses on job priority, not resource guarantees per tenant.

4. Final Answer:

   Capacity Scheduler with minimum and maximum queue capacities set -> Option D

5. Quick Check:

   Capacity Scheduler = guaranteed min + flexible max [OK]

Hint: Capacity Scheduler supports min and max resource guarantees [OK]

Common Mistakes:

• Choosing FIFO for guaranteed resources
• Ignoring min capacity in Fair Scheduler
• Confusing priority with resource guarantees