MongoDB_Spark

Apache Spark and MongoDB
Turning Analytics into Real-Time Action
A MongoDB White Paper

Table of Contents
1Introduction
2Rich Analytics in MongoDB
3Apache Spark: Extending MongoDB Analytics
5Configuring & Running Spark with MongoDB
7Integrating MongoDB & Spark with BI & Hadoop
10MongoDB & Spark in the Wild
10Conclusion
10We Can Help

Introduction
We live in a world of “big data”. But it isn’t only the data
itself that is valuable – it is the insight it can generate. That
insight can help designers better predict new products that
customers will love. It can help manufacturing companies
model failures in critical components, before costly recalls.
It can help financial institutions detect fraud. It can help
retailers better forecast supply and demand. Marketing
departments can deliver more relevant recommendations
to its customers. The list goes on.
How quickly an organization can unlock and act on that
insight is becoming a major source of competitive
advantage. Collecting data in operational systems and then
relying on nightly batch ETL (Extract Transform Load)
processes to update the Enterprise Data Warehouse
(EDW) is no longer sufficient. Speed-to-insight is critical,
and so analytics against live operational data to drive
real-time action is fast becoming a necessity, enabled by
technologies like MongoDB and Apache Spark.
Why Apache Spark and MongoDB?
Apache Spark is one of the fastest-growing big data
projects in the history of the Apache Software Foundation.
With its memory-oriented architecture, flexible processing
libraries and ease-of-use, Spark has emerged as a leading
distributed computing framework for real-time analytics.
With over 10 million downloads, MongoDB is the most
popular non-relational database, counting more than one
third of the Fortune 100 as customers. Its flexible
JSON-based document data model, dynamic schema and
automatic scaling on commodity hardware make MongoDB
an ideal fit for modern, always-on applications that must
manage high volumes of rapidly changing, multi-structured
data. Internet of Things (IoT), mobile apps, social
engagement, customer data and content management
systems are prime examples of MongoDB use cases.
Combining the leading analytics processing engine with
the fastest-growing database enables organizations to
realize real time analytics. Spark jobs can be executed
directly against operational data managed by MongoDB
without the time and expense of ETL processes. MongoDB
1

can then efficiently index and serve analytics results back
into live, operational processes. This approach offers many
benefits to teams tasked with delivering modern, data
driven applications:
• Developers can build more functional applications
faster, using a single database technology.
• Operations teams eliminate the requirement for
shuttling data between separate operational and
analytics infrastructure, each with its own unique
configuration, maintenance and management
requirements.
• CIOs deliver faster time-to-insight for the business, with
lower cost and risk.
This whitepaper will discuss the analytics capabilities
offered by MongoDB and Apache Spark, before providing
an overview of how to configure and combine them into a
real-time analytics engine. It will conclude with example
use cases.
Rich Analytics in MongoDB
Unlike NoSQL databases that offer little more than basic
key-value query operations, developers and data scientists
can use MongoDB’s native query processing and data
navigation capabilities to generate many classes of
analytics, before needing to adopt dedicated frameworks
such as Spark or Hadoop for more specialized tasks.
Leading organizations including Salesforce, McAfee,
Buzzfeed, Intuit, City of Chicago, Amadeus, Buffer and
many others rely on MongoDB’s powerful query
functionality, aggregations and indexing to generate
analytics in real-time directly against their live, operational
data.
In this section of the paper, we provide more detail on
MongoDB's native analytics features, before then exploring
the additional capabilities Apache Spark can bring.
MongoDB Query Model
MongoDB users have access to a broad array of query,
projection and update operators supporting analytical
queries against live operational data:
• Aggregation and MapReduce queries, discussed in
more detail below
• Range queries returning results based on values
defined as inequalities (e.g., greater than, less than or
equal to, between)
• Geospatial queries returning results based on proximity
criteria, intersection and inclusion as specified by a
point, line, circle or polygon
• Text search queries returning results in relevance order
based on text arguments using Boolean operators (e.g.,
AND, OR, NOT)
• Key-value queries returning results based on any field in
the document, often the primary key
• Native BI and Hadoop integration, for deep, offline
analytics.
With the combination of MongoDB’s dynamic document
model and comprehensive query framework, users are able
to store data before knowing all of the questions they will
need to ask of it.
Data Aggregation
The MongoDB Aggregation Pipeline is similar in concept to
the SQL GROUP BY statement, enabling users to
generate aggregations of values returned by the query
(e.g., count, minimum, maximum, average, intersections)
that can be used to power analytics dashboards and
vizualizations.
Using the Aggregation Framework, documents in a
MongoDB collection (analogous to a table in a relational
database) pass through an aggregation pipeline, where
they are processed by operators. Expressions produce
output documents based on calculations performed on the
input documents. The accumulator expressions used in the
$group operator maintain state (e.g., totals, mins, maxs,
averages, standard deviations) as documents progress
through the pipeline. MongoDB also includes the ability to
combine data from multiple collections by implementing
left outer joins through the $lookup operator, which can
now be included as an aggregation stage.
The aggregation pipeline enables multi-step data
enrichment and transformations to be performed directly in
2

the database with a simple declarative interface, supporting
processes such as lightweight ETL to be performed within
MongoDB.
Result sets from the aggregation pipeline can be written to
a named collection with no limit to the output size (subject
to the underlying storage system). Existing collections can
be replaced with new results while maintaining previously
defined indexes to ensure queries can always be returned
efficiently over rapidly changing data.
Incremental MapReduce within MongoDB
MongoDB provides native support for MapReduce,
enabling complex data processing that is expressed in
JavaScript and executed across data in the database.
Multiple MapReduce jobs can be executed concurrently
across both single servers and sharded collections.
The incremental MapReduce operation uses a temporary
collection during processing so it can be run periodically
over the same target collection without affecting
intermediate states. This mode is useful when continuously
updating statistical output collections used by reporting
dashboards.
Optimizing Analytics Queries with
MongoDB Indexes
MongoDB provides a number of different index types to
optimize performance of real-time and ad-hoc analytics
queries across highly variable, fast moving data sets. These
same indexes can be used by Apache spark to filter only
interesting subsets of data against which analytics can be
run.
Indexes can be created on any field within a document or
sub-document. In addition to supporting single field and
compound indexes, MongoDB also supports indexes of
arrays with multiple values, short-lived data (i.e., Time To
Live), partial, sparse, geospatial and text data, and can
enforce constraints with unique indexes. Index intersection
allows indexes to be dynamically combined at runtime to
efficiently answer explorative questions of the data. Refer
to the documentation for the full list of index types.
The MongoDB query optimizer selects the best index to
use by running alternate query plans and selecting the
index with the best response time for each query type. The
results of this empirical test are stored as a cached query
plan and are updated periodically.
MongoDB also supports covered queries – where returned
results contain only indexed fields, without having to
access and read from source documents. With the
appropriate indexes, analytics workloads can be optimized
to use predominantly covered queries.
Building upon MongoDB, Apache Spark can offer
additional analytics capabilities to serve real time,
operational processes.
Apache Spark: Extending
MongoDB’s Analytics
Capabilities
Apache Spark is a powerful open source processing
engine designed around speed, ease of use, and
sophisticated analytics. Originally developed at UC
Berkeley in 2009, Spark has seen rapid adoption by
enterprises across a range of industries.
Spark is a general-purpose framework used for many types
of data processing. Spark comes packaged with support
for machine learning, interactive queries (SQL), statistical
queries with R, graph processing, ETL, and streaming. For
loading and storing data, Spark integrates with a number of
storage and messaging systems including Amazon S3,
Kafka, HDFS, RDBMSs, and MongoDB. Additionally, Spark
supports a variety of popular development languages
including Java, Python and Scala.
Apache Spark Benefits
Spark was initially designed for interactive queries and
iterative algorithms, as these were two major use cases not
well served by batch frameworks such as Hadoop’s
MapReduce. Consequently Spark excels in scenarios that
require fast performance, such as iterative processing,
interactive querying, large-scale batch operations,
streaming, and graph computations.
Developers and data scientists typically deploy Spark for:
3

FigurFigure 1:e 1: Apache Spark Ecosystem
• Simplicity. Easy-to-use APIs for operating on large
datasets. This includes a collection of sophisticated
operators for transforming and manipulating
semi-structured data.
• Speed. By exploiting in-memory optimizations, Spark
has shown up to 100x higher performance than
MapReduce running on Hadoop.
• Unified Framework. Packaged with higher-level libraries,
including support for SQL queries, machine learning,
stream and graph processing. These standard libraries
increase developer productivity and can be combined to
create complex workflows.
Spark allows programmers to develop complex, multi-step
data pipelines using a directed acyclic graph (DAG)
pattern. It also supports in-memory data sharing across
DAGs, so that different jobs can work with the same data.
Spark takes MapReduce to the next level with less
expensive shuffles during data processing. Spark holds
intermediate results in memory rather than writing them to
disk, which is useful especially when a process needs to
iteratively work on the same dataset. Spark is designed as
an execution engine that works with data both in-memory
and on-disk. Spark operators perform external operations
when data does not fit in memory. Spark can be used for
processing datasets that are larger than the aggregate
memory in a cluster.
The Resilient Distributed Dataset (RDD) is the core data
structure used in the Spark framework. RDD is analogous
to a table in a relational database or a collection in
MongoDB. RDDs are immutable – they can be modified
with a transformation, but the transformation returns a new
RDD while the original RDD remains unchanged.
RDDs supports two types of operations: Transformations
and Actions. Transformations return a new RDD after
processing it with a function such as map, filter, flatMap,
groupByKey, reduceByKey, aggregateByKey, pipe, or
coalesce.
The Action operation evaluates the RDD and returns a new
value. When an Action function is called on a RDD object,
all the processing queries are computed and the result
value is returned. Some of the Action operations are
reduce, collect, count, first, take, countByKey, and foreach.
When to Use Spark with MongoDB
While MongoDB natively offers rich analytics capabilities,
there are use-cases where integrating the Spark engine
can extend the real-time processing of operational data
managed by MongoDB, and allow users to operationalize
results generated from Spark within real-time business
processes supported by MongoDB.
Spark can take advantage of MongoDB’s rich secondary
indexes to extract and process only the range data it needs
– for example, analyzing all customers located in a specific
geography. This is very different from other databases that
either do not offer, or do not recommend the use of
secondary indexes. In these cases, Spark would need to
extract all data based on a simple primary key, even if only
a subset of that data is required for the Spark process. This
means more processing overhead, more hardware, and
longer time-to-insight for the analyst.
Examples of where it is useful to combine Spark and
MongoDB include the following.
Rich Operators & Algorithms
Spark supports over 100 different operators and
algorithms for processing data. Developers can use these
to perform advanced computations that would otherwise
require more programmatic effort combining the MongoDB
aggregation framework with application code. For example,
Spark offers native support for advanced machine learning
algorithms including k-means clustering and Gaussian
mixture models.
Consider a web analytics platform that uses the MongoDB
aggregation framework to maintain a real time dashboard
displaying the number of clicks on an article by country;
how often the article is shared across social media; and the
number of shares by platform. With this data, analysts can
4

quickly gain insight on how content is performing,
optimizing user’s experience for posts that are trending,
with the ability to deliver critical feedback to the editors and
ad-tech team.
Spark’s machine learning algorithms can also be applied to
the log, clickstream and user data stored in MongoDB to
build precisely targeted content recommendations for its
readers. Multi-class classifications are run to divide articles
into granular sub-categories, before applying logistic
regression and decision tree methods to match readers’
interests to specific articles. The recommendations are
then served back to users through MongoDB, as they
browse the site.
Processing Paradigm
Many programming languages can use their own
MongoDB drivers to execute queries against the database,
returning results to the application where additional
analytics can be run using standard machine learning and
statistics libraries. For example, a developer could use the
MongoDB Python or R drivers to query the database,
loading the result sets into the application tier for additional
processing.
However, this starts to become more complex when an
analytical job in the application needs to be distributed
across multiple threads and nodes. While MongoDB can
service thousands of connections in parallel, the application
would need to partition the data, distribute the processing
across the cluster, and then merge results. Spark makes
this kind of distributed processing easier and faster to
develop. MongoDB exposes operational data to Spark’s
distributed processing layer to provide fast, real-time
analysis. Combining Spark queries with MongoDB indexes
allows data to be filtered, avoiding full collection scans and
delivering low-latency responsiveness with minimal
hardware and database overhead.
Skills Re-Use
With libraries for SQL, machine learning and others –
combined with programming in Java, Scala and Python –
developers can leverage existing skills and best practices
to build sophisticated analytics workflows on top of
MongoDB.
Configuring & Running Spark
with MongoDB
There are currently two packaged connectors to integrate
MongoDB and Spark:
• The MongoDB Connector for Hadoop and Spark
• The Spark-MongoDB Connector, developed by Stratio
MongoDB Connector for Hadoop & Spark
The MongoDB Connector is a plugin for both Hadoop and
Spark that provides the ability to use MongoDB as an input
source and/or an output destination for jobs running in
both environments. Note that the connector directly
integrates Spark with MongoDB and has no dependency
on also having a Hadoop cluster running.
Input and output classes are provided allowing users to
read and write against both live MongoDB collections and
against BSON (Binary JSON) files that are used to store
MongoDB snapshots. Standard MongoDB connection
strings, including authentication credentials and read
preferences, are used to specify collections against which
Spark queries are run, and where results are written back
to. JSON formatted queries and projections can be used to
filter the input collection, which uses a method in the
connector to create a Spark RDD from the MongoDB
collection.
There are only two dependencies for installing the
connector:
• Download the MongoDB Connector for Hadoop and
Spark. For Spark, all you need is the "core" jar.
• Download the jar for the MongoDB Java Driver.
The documentation provides a worked example using
MongoDB as both a source and sink for Spark jobs,
including how to create and save Spark’s Resilient
Distributed Dataset (RDD) to MongoDB collections.
Databricks (founded by the creators of Spark) and
MongoDB have also collaborated in creating a
demonstration app that documents how to configure and
5

MongoDB Connector for HadoopMongoDB Connector for Hadoop Stratio Spark-MongoDB ConnectorStratio Spark-MongoDB Connector
Machine Learning Yes Yes
SQL Not currently Yes
DataFrames Not currently Yes
Streaming Not currently Not currently
Python Yes Yes
Using SparkSQL syntax
Use MongoDB secondary indexes
to filter input data
Yes Yes
Compatibility with MongoDB
replica sets and sharding
Yes Yes
MongoDB Support Yes
Read and write
Yes
Read and write
HDFS Support Yes
Read and write
Partial
Write only
Support for MongoDB BSON
Files
Yes No
Commercial Support Yes
With MongoDB Enterprise Advanced
Yes
Provided by Stratio
TTable 1:able 1: Comparing Spark Connectors for MongoDB
use Spark to execute analysis of market trading data
managed by MongoDB.
Spark-MongoDB Connector
The Spark-MongoDB Connector is a library that allows the
user to read and write data to MongoDB with Spark,
accessible from Python, Scala and Java API’s. The
Connector is developed by Stratio and distributed under
the Apache Software License.
The MongoDB SparkSQL data source implements the
Spark Dataframe API, and is fully implemented in Scala.
The connector allows integration between multiple data
sources that implement the same API for Spark, in addition
to using the SQL syntax to provide higher-level
abstractions for complex analytics. It also includes an easy
way to integrate MongoDB with Spark Python API.
Stratio's SparkSQL MongoDB connector implements the
PrunedFilteredScan API instead of the TableScan API.
Therefore, in addition to MongoDB’s query projections, the
connector can avoid scanning the entire collection, which is
not the case with other databases.
The connector supports the Spark Catalyst optimizer for
both rule-based and cost-based query optimization. To
operate against multi-structured data, the connector infers
the schema by sampling documents from the MongoDB
collection. This process is controlled by the samplingRatio
parameter. If the schema is known, the developer can
provide it to the connector, avoiding the need for any
inference. Once data is stored in MongoDB, Stratio
provides an ODBC/JDBC connector for integrating results
with any BI tool, such as Tableau and QlikView.
The connector can be downloaded from the community
Spark Packages repository.
Installation is simple – the connector can be included in a
Spark application with a single command.
6

FigurFigure 2:e 2: Using MongoDB Replica Sets to Isolate
Analytics from Operational Workloads
You can learn more about the Spark-Mongodb connector
from Stratio’s GitHub page.
Operations: Separating Operational from
Analytical Workloads
Spark can be run on any physical node within a MongoDB
replica set. However, executing heavyweight analytics jobs
on a host that is also serving requests from an operational
application can cause contention for system resources.
Similarly the working set containing the indexes and most
frequently accessed data in memory is overwritten by data
needed to process analytics queries. Whether running
MongoDB aggregations or Spark processes, it is a best
practice to isolate competing traffic from the operational
and analytics workloads.
Using native replication, MongoDB maintains multiple
copies of data called replica sets. Replica sets are primarily
designed to provide high availability and disaster recovery,
and to isolate different workloads running over the same
data set.
By default, all read and write operations are routed to the
primary replica set member, thus ensuring strong
consistency of data being read and written to the database.
Applications can also be configured to read from
secondary replica set members, where data is eventually
consistent by default. Reads from secondaries can be
useful in scenarios where it is acceptable for data to be
several seconds (typically milliseconds) behind the live
data, which is typical for many analytics and reporting
applications. To ensure a secondary node serving analytics
queries is never considered for election to the primary
replica set member, it should be configured with a priority
of 0.
Unlike other databases that have to offload Spark to
dedicated analytics nodes that are running an entirely
different configuration from the database nodes, all
members of a MongoDB replica set share the same
availability, scalability and security mechanisms. This
approach reduces operational complexity and risk.
Integrating MongoDB & Spark
Analytics with BI Tools &
Hadoop
BI Integration
Real time analytics generated by MongoDB and Spark can
serve both online operational applications and offline
reporting systems, where it can be blended with historical
data and analytics from other data sources. To power
dashboards, reports and visualizations data stored in
MongoDB can be easily explored with industry-standard
SQL-based BI and analytics platforms. Using the
MongoDB Connector for BI, analysts, data scientists and
business users can seamlessly visualize semi-structured
and unstructured data managed in MongoDB, alongside
traditional data in their SQL databases, using the same BI
tools deployed within millions of enterprises.
The Connector for BI translates SQL statements issued by
the BI tool into equivalent MongoDB queries that are then
sent to MongoDB for processing, and then returns results
for visualization in the tool’s UI.
Hadoop Integration
Like MongoDB, Hadoop is seeing growing adoption across
industry and government, becoming an adjunct to – and in
7

some cases a replacement of – the traditional Enterprise
Data Warehouse.
Many organizations are harnessing the power of Hadoop
and MongoDB together to create complete big data
applications:
• MongoDB powers the online, real time operational
application, serving business processes and end-users
• Hadoop consumes data from MongoDB, blending its
with data from other operational systems to fuel
sophisticated analytics and machine learning. Results
are loaded back to MongoDB to serve smarter
operational processes – i.e., delivering more relevant
offers, faster identification of fraud, improved customer
experience.
Organizations such as eBay, Orbitz, Pearson, Square Enix
and SFR are using MongoDB alongside Hadoop to
operationalise big data.
As well as providing integration for Spark, the MongoDB
Connector for Hadoop and Spark is also the foundation for
MongoDB’s integration with Hadoop. A single extensible
connector supporting both Spark and Hadoop makes it
FigurFigure 3:e 3: Integrating MongoDB & Spark with BI Tools
much simpler for developers and operations teams to build
and run powerful big data pipelines.
The MongoDB Connector for Hadoop and Spark presents
MongoDB as a Hadoop data source allowing Hadoop jobs
to read data from MongoDB directly without first copying it
to HDFS, thereby eliminating the need to move TB of data
between systems. Hadoop jobs can pass queries as filters,
thereby avoiding the need to scan entire collections and
speeding up processing; they can also take advantage of
MongoDB’s indexing capabilities, including text and
geospatial indexes.
As well as reading from MongoDB, the Connector also
allows results of Hadoop jobs to be written back out to
MongoDB, to support real-time operational processes and
ad-hoc querying by both MongoDB and Spark query
engines.
The Connector supports MapReduce, Spark on HDFS, Pig,
Hadoop Streaming (with Node.js, Python or Ruby) and
Flume jobs. Support is also available for SQL queries from
Apache Hive to be run across MongoDB data sets. The
connector is certified on the leading Hadoop distributions
from Cloudera, Hortonworks and MapR.
8

To learn more about running MongoDB with Hadoop,
download the Big Data: Examples and Guidelines
whitepaper
MongoDB & Spark in the Wild
Many organizations are already combining MongoDB and
Spark to unlock real-time analytics from their operational
data:
• A global manufacturing company has built a pilot project
to estimate warranty returns by analyzing material
samples from production lines. The collected data
enables them to build predictive failure models using
Spark Machine Learning and MongoDB.
• A video sharing website is using Spark with MongoDB
to place relevant advertisements in front of users as
they browse, view and share videos.
A multinational banking group operating in 31 countries
with 51 million clients implemented a unified real-time
monitoring application with the Stratio Big Data (BD)
platform, running Apache Spark and MongoDB. The bank
wanted to ensure a high quality of service across its online
channels, and so needed to continuously monitor client
activity to check service response times and identify
potential issues. The technology stack included:
• Apache Flume to aggregate log data
• Apache Spark to process log events in real time
• MongoDB to persist log data, processed events and
Key Performance Indicators (KPIs).
The aggregated KPIs enable the bank to meet its objective
of analyzing client and systems behavior in order to
improve customer experience. Collecting raw log data
allows the bank to immediately rebuild user sessions if a
service fails, with the analysis providing complete
traceability to identify the root cause of the issue.
MongoDB was selected as the database layer of the
application because of it’s always-on availability, high
performance and linear scalability. The application needed
a database with a fully dynamic schema to support high
volumes of rapidly changing semi-structured and
unstructured data being ingested from the variety of logs,
clickstreams and social networks. Operational efficiency
was also important. The bank wanted to automate
orchestrating the application flow and have deep visibility
into the database with proactive monitoring. MongoDB
Cloud Manager provided the required management
platform.
Conclusion
MongoDB natively provides a rich analytics framework
within the database. Multiple connectors are also available
to integrate Spark with MongoDB to enrich analytics
capabilities by enabling analysts to apply libraries for
machine learning, streaming and SQL to MongoDB data.
Together MongoDB and Apache Spark are already
enabling developers and data scientists to turn analytics
into real-time action.
We Can Help
We are the MongoDB experts. Over 2,000 organizations
rely on our commercial products, including startups and
more than a third of the Fortune 100. We offer software
and services to make your life easier:
MongoDB Enterprise Advanced is the best way to run
MongoDB in your data center. It’s a finely-tuned package
of advanced software, support, certifications, and other
services designed for the way you do business.
MongoDB Cloud Manager is the easiest way to run
MongoDB in the cloud. It makes MongoDB the system you
worry about the least and like managing the most.
MongoDB Professional helps you manage your
deployment and keep it running smoothly. It includes
support from MongoDB engineers, as well as access to
MongoDB Cloud Manager.
Development Support helps you get up and running quickly.
It gives you a complete package of software and services
for the early stages of your project.
9

MongoDB Consulting packages get you to production
faster, help you tune performance in production, help you
scale, and free you up to focus on your next release.
MongoDB Training helps you become a MongoDB expert,
from design to operating mission-critical systems at scale.
Whether you’re a developer, DBA, or architect, we can
make you better at MongoDB.
Contact us to learn more, or visit www.mongodb.com.
Resources
For more information, please visit mongodb.com or contact
us at sales@mongodb.com.
Case Studies (mongodb.com/customers)
Presentations (mongodb.com/presentations)
Free Online Training (university.mongodb.com)
Webinars and Events (mongodb.com/events)
Documentation (docs.mongodb.org)
MongoDB Enterprise Download (mongodb.com/download)
New York • Palo Alto • Washington, D.C. • London • Dublin • Barcelona • Sydney • Tel Aviv
US 866-237-8815 • INTL +1-650-440-4474 • info@mongodb.com
© 2015 MongoDB, Inc. All rights reserved.
10

MongoDB_Spark

More Related Content

What's hot (20)

Similar to MongoDB_Spark (20)

More from Mat Keep (11)

MongoDB_Spark