All MCQs

Failures and speculation can cause task attempts to repeat.
Idempotence prevents duplicate side effects from corrupting results.

Disk I/O and repeated job startup increase latency.
In-memory engines such as Spark can be faster for iterative pipelines.

Complex workflows are decomposed into stages.
Intermediate HDFS output provides durability between jobs.

Samples approximate the key distribution.
Good boundaries balance reducer loads while preserving global ranges.

Range partitioning ensures reducer outputs cover consecutive key intervals.
Concatenating part files in reducer order yields global order.

Each mapper keeps only its best N records.
The final reducer processes at most roughly N candidates per mapper.

Local top-N lists can be produced by mappers and merged globally.
This reduces the amount of data sent to the final stage.

The shuffle groups duplicate occurrences under the same key.
A reducer can emit one result per distinct group.

Mappers emit each word with count one.
Reducers sum the counts for identical words.

Mappers emit terms with document identifiers.
Reducers aggregate the document lists for each term.

Filters are often implemented as map-only jobs.
Records that fail the predicate produce no output.

Identity reduction can preserve framework grouping and sorting results.
It may also serve as a simple pass-through stage.