All MCQs

The ResourceManager tracks application attempts.
A replacement ApplicationMaster can recover or restart work depending on framework support.

Spark, Tez, and other engines can request containers through YARN.
Storage and resource management are therefore decoupled from MapReduce.

Priority can guide scheduling among otherwise eligible applications.
Queue policies and fairness rules still constrain allocation.

Rack-local access crosses a node network link but not the rack switch boundary.
It is less ideal than node-local but better than off-rack.

Node locality provides the lowest network cost for reading HDFS input.
It is preferred when resources are available.

Local or rack-local execution reduces network traffic.
The scheduler may balance locality against waiting time.

For example, nodes may have spare memory but insufficient CPU for a requested container.
Better sizing and scheduling can improve packing.

A request must fit the available resources on a node.
Large indivisible requests can wait while smaller resource fragments remain unused.

Preemption enforces queue guarantees or priorities.
It can reduce efficiency if containers are killed frequently.

Queues organize users, teams, or workload classes.
Policies can set capacity, priority, and access controls.

When one application is alone, it may use extra capacity.
As others arrive, resources move toward fair shares.

Queues receive guaranteed capacity while unused resources may be shared.
This supports predictable multi-tenant cluster use.