Wednesday, 17 November 2021

Top 20 Apache Spark Interview Questions & Answers


Ques: 1). What is Apache Spark?


Apache Spark is an open-source real-time processing cluster computing framework. It has a vibrant open-source community and is now the most active Apache project. Spark is a programming language that allows you to programme large clusters with implicit data parallelism and fault tolerance.

Spark is one of the Apache Software Foundation's most successful projects. Spark has unquestionably established itself as the market leader in Big Data processing. Spark is used by many enterprises on clusters with thousands of nodes. Spark is now used by large companies like as Amazon, eBay, and Yahoo!

Ques: 2). What advantages does Spark have over MapReduce?


Compared to MapReduce, Spark has the following advantages:

Spark implements processing 10 to 100 times quicker than Hadoop MapReduce due to the availability of in-memory processing, whereas MapReduce uses persistence storage for any of the data processing activities.

Unlike Hadoop, Spark has built-in libraries that allow it to do a variety of functions from the same core, including as batch processing, steaming, machine learning, and interactive SQL queries. Hadoop, on the other hand, only supports batch processing.

Hadoop is heavily reliant on discs, but Spark encourages caching and data storage in memory. Spark is capable of performing computations multiple times on the same dataset. This is called iterative computation while there is no iterative computing implemented by Hadoop.


Ques: 3). What exactly is YARN?


YARN is a fundamental element in Spark, similar to Hadoop, in that it provides a central and resource management platform for delivering scalable operations across the cluster. YARN, like Mesos, is a distributed container manager, but Spark is a data processing tool. Spark can be run on YARN in the same way that Hadoop Map Reduce can. Running Spark on YARN needs the use of a Spark binary distribution with YARN support.


Ques: 4). What's the difference between an RDD, a Dataframe, and a Dataset?


Resilient Distributed Dataset (RDD) - RDD stands for Resilient Distributed Dataset. It is the most basic data structure in Spark, consisting of an immutable collection of records partitioned among cluster nodes. It allows us to do fault-tolerant in-memory calculations on massive clusters.

RDD, unlike DF and DS, will not keep the schema. It merely stores data. If a user wants to apply a schema to an RDD, they must first build a case class and then apply the schema to the data.

We will use RDD for the below cases:

-When our data is unstructured, A streams of text or media streams.

-When we don’t want to implement any schema.

-When we don’t care about the column name attributes while processing or accessing.

-When we want to manipulate the data with functional programming constructs than domain specific expressions.

-When we want low-level transformation, actions and control on the dataset.


-Like RDD DataFrames are immutable collection of data.

-Unlike RDD DataFrame will have schema for their data making user to easily access/process large set of data which is distributed among the nodes of cluster.

-DataFrame provides a domain specific language API to manipulate distributed data and makes Spark accessible to a wider audience, beyond specialized data engineers.

-From Spark 2.x Spark DataFrames are nothing but Dataset[Row] or alias (Untyped API)

consider DataFrame as an alias for a collection of generic objects Dataset[Row], where a Row is a generic untyped JVM object


Dataset, by contrast, is a collection of strongly-typed JVM objects, dictated by a case class you define in Scala or a class in Java. To implement case class on RDD and use as Dataset[T].


Ques: 5). Can you explain how you can use Apache Spark along with Hadoop?


Apache Spark provides the benefit of being Hadoop compatible. They make a powerful tech team when they work together. Using Apache Spark and Hadoop combines the processing capability of Spark with the best capabilities of HDFS and YARN. The following are some examples of how to use Hadoop Components with Apache Spark:

Batch & Real-Time Processing – MapReduce and Spark can work together, where the former handles batch processing, and the latter handles real-time processing.

HDFS – Spark can make use of the HDFS to leverage the distributed replicated storage.

MapReduce – Apache Spark can be used in conjunction with MapReduce in a similar Hadoop cluster or independently as a processing framework.

YARN – You can run Spark applications on YARN.


Ques: 6). What is the meaning of the term "spark lineage"?


• In Spark, regardless of the actual data, all dependencies between RDDs will be logged in a graph. In Spark, this is referred to as a lineage graph.

• RDD stands for Resilient Distributed Dataset, with the term "resilient" referring to fault tolerance. We can re-compute the missing or damaged partition due to node failure using RDD Lineage Graph. When we generate new RDDs based on existing RDDs, we use lineage graph spark to handle the dependencies. Each RDD keeps a pointer to one or more parents, together with metadata describing the type of relationship it has with the parent RDD.

• RDD Lineage Graph in Spark can be obtained using the ToDebugString method.


Ques: 7). List the various components of the Spark Ecosystem.


These are the five types of components in the Spark Ecosystem:

GraphX: Enables graphs and graph-parallel computation.

MLib: It is used for machine learning.

Spark Core: A powerful parallel and distributed processing platform.

Spark Streaming: Handles real-time streaming data.

Spark SQL: Combines Spark's functional programming API with relational processing.


Ques: 8). What is RDD in Spark? Write about it and explain it.


Resilient Distributed Dataset (RDD) is an acronym for Resilient Distributed Dataset. RDDs are a fault-tolerant core data structure in Spark that is immutable. They've disseminated partitioned datasets among the cluster nodes.

Parallelizing and referencing a data set are the two methods for constructing RDDS. Lazy evaluation is the responsibility of the RDDS. The faster processing performance in Spark is due to the lazy evaluation of RDDs.


Ques: 9). In Spark, how does streaming work?


Spark gets data in real time that is separated into batches. The Spark Engine processes these batches of data, and the final stream of results is returned back in batches. DStream, or Discretized Stream, is the most basic stream unit in Spark.


Ques: 10). Is it feasible to access and analyse data stored in Cassandra databases using Apache Spark?


Yes, Apache Spark may be used to retrieve and analyse data stored in Cassandra databases. Apache Spark can access and analyse data contained in Cassandra databases using the Spark Cassandra Connector. Spark should have a functionality that allows Spark executors to communicate with local Cassandra nodes and request just local data.

Cassandra and Apache Spark can be connected to speed up queries by lowering network traffic between Spark executors and Cassandra nodes.


Ques: 11). What are the advantages of using Spark SQL?


Spark SQL carries out the following tasks:

Loads data from a variety of structured datasources, such as a relational database management system (RDBMS).

It may query data using SQL commands within the Spark programme as well as JDBC/ODBC connectors from third-party tools such as Tableau.

It can also provide SQL and Python/Scala code interaction.


Ques: 12). What is the purpose of Spark Executor?


The Executors are obtained on top of worker nodes in the clusters when a SparkContext is formed. Spark Executors are in charge of performing computations and storing data on the worker node. They are also in charge of returning the results to the driver.


Ques: 13). What are the advantages and disadvantages of Spark?


Advantages: Spark is known for real-time data processing, which may be employed in applications such as stock market analysis, finance, and telecommunications.

Spark's stream processing allows for real-time data analysis, which can aid in fraud detection, system alarms, and other applications.

Due to its lazy evaluation mechanism and parallel processing, Spark processes data 10 to 100 times quicker.

Disadvantages: When compared to Hadoop, Spark consumes greater storage space.

The task is distributed over numerous clusters rather than taking place on a single node.

Spark's in-memory processing might be costly when dealing with large amounts of data.

When compared to Hadoop, Spark makes better use of data.


Ques: 14). What are some of the drawbacks of utilising Apache Spark?


The following are some of the drawbacks of utilising Apache Spark:

There is no file management system built-in. To take benefit of a file management system, integration with other platforms such as Hadoop is essential.

Higher latency, but lower throughput as a result

It does not support the processing of real-time data streams. In Apache Spark, live data streams are partitioned into batches, which are then processed and turned back into batches. To put it another way, Spark Streaming is more like micro-batch data processing than true real-time data processing.

There are fewer algorithms available.

Record-based window requirements are not supported by Spark streaming. It is necessary to distribute work across multiple clusters instead of running everything on a single node.

Apache Spark's in-memory ability becomes a bottleneck when used for the cost-efficient processing of big data.


Ques: 15). Is Apache Spark compatible with Apache Mesos?


Yes. Spark can work on Apache Mesos-managed clusters, just as it works on YARN-managed clusters. Spark may run without a resource manager in standalone mode. If it has to execute on multiple nodes, it can use YARN or Mesos.


Ques: 16). What are broadcast variables, and how do they work?


Accumulators and broadcast variables are the two types of shared variables in Spark. Instead of shipping back and forth to the driver, the broadcast variables are read-only variables cached in the Executors for local referencing. A broadcast variable preserves a read-only cached version of a variable on each computer instead of delivering a copy of the variable with tasks.

Additionally, broadcast variables are utilised to distribute a copy of a big input dataset to each node. To cut transmission costs, Apache Spark distributes broadcast variables using efficient broadcast algorithms.

There is no need to replicate variables for each task when using broadcast variables. As a result, data can be processed quickly. In contrast to RDD lookup(), broadcast variables assist in storing a lookup table inside the memory, enhancing retrieval efficiency.


Ques: 17). In Apache Spark, how does caching work?


Caching RDDs in Spark speeds up processing by allowing numerous accesses to the same RDD. The function of Discretized Streams, or DStreams, in Spark streaming is to allow users to cache or retain data in memory.

The functions cache () and persist(level) are used to cache data in memory and cache memory based on the storage level specified, respectively.

The persist () without the level specifier is the same as cache, which means it caches the data in memory. The persist(level) method caches data at the provided storage level, such as on disc, on the server, or in off-heap memory.


Ques: 18).  What exactly is Akka? What does Spark do with it?


Akka is a Scala and Java framework for reactive, distributed, parallel, and robust concurrent applications. Akka is the foundation for Apache Spark.

When assigning tasks to worker nodes, Spark employs Akka for job scheduling and messaging between the master and the worker node.


Ques: 19). What applications do you utilise Spark streaming for?


When real-time data must be streamed into the Spark programme, this method is employed. It can be broadcast from a variety of places, such as Kafka, Flume, Amazon Kinesis, and others. For processing, the streamed data is separated into batches.

Spark streaming is used to conduct real-time sentiment analysis of customers on social media sites like as Twitter and Facebook, among others.

Live streaming data processing is critical for detecting outages, detecting fraud in financial institutions, and making stock market predictions, among other things.


Ques: 20). What exactly do you mean when you say "lazy evaluation"?


The way Spark works with data is intellectual. When you ask Spark to perform a task on a dataset, it follows your instructions and records them so that it doesn't forget them - but it doesn't do anything until you tell it to. When map() is invoked on an RDD, the operation is not done immediately. Transformations aren't evaluated by Spark until you use them. This aids in the overall data processing workflow optimization.

No comments:

Post a Comment