Getting started with AWS IoT and a dockerized device

IoT isn’t a new term and has, actually, been one of the buzz words of the XXI century, right alongside Crypto Currency. The two of them are seen holding hand in hand in the good and the bad, from interesting crypto projects focused on IoT usage, or actual IoT devices being hacked to mine crypto currencies. But this article is actually about IoT: it seems the last few years have given us the perfect momentum between three main ingredients for its popularity: the exponential increase of network capable devices, the easily available processing power and, finally, the hungry-for-data dynamic we from most parties nowadays.

For someone working as a Data Engineer (or related) there isn’t there much of a more end-to-end project than one which goes from setting up an actual no-so-intelligence device, design the methods to connect these devices to a network and, still, go about designing all streaming/batch processing infrastructure needed to deal with all the data. It really is a huge challenge.

In the following lines, I’ll focus on how AWS IoT has come to help in the first and second challenge and show an example of how to emulate an actual IoT device with the help of docker and walk through some of the nice and easy features AWS IoT offers such as IoT Core, management, topics and rules as well as integration with other AWS services such as Kinesis, Firehose or Lambdas.

Continue reading “Getting started with AWS IoT and a dockerized device”

Build a Data Pipeline with AWS Athena and Airflow (part 1)

In this post, I build up on the knowledge shared in the post for creating Data Pipelines on Airflow and introduce new technologies that help in the Extraction part of the process with cost and performance in mind. I’ll go through the options available and then introduce to a specific solution using AWS Athena. First we’ll establish the dataset and organize our data in S3 Buckets. Afterwards, you’ll learn how to make it so that this information is queryable through AWS Athena, while making sure it is updated daily.

Data dump files of not so structured data are a common byproduct of Data Pipelines that include extraction. dumps of not-so-structured data. This happens by design: business-wise and as Data Engineers, it’s never too much data. From an investment stand point, object-relational database systems can become increasingly costly to keep, especially if we aim at keeping performance while the data grows.

Having this said, this is not a new problem. Both Apache and Facebook have developed open source software that is extremely efficient in dealing with extreme amounts of data. While such softwares are written in Java, they maintain an abstracted interface to the data that relies on traditional SQL language to query data that is stored on filesystem storage, such as S3 for our example and in a wide range of different formats from HFDS to CSV.

Today we have many options to tackle this problem and I’m going to go through on how to welcome this problem in today’s serverless world with AWS Athena. For this we need to quickly rewind back in time and go through the technology Continue reading “Build a Data Pipeline with AWS Athena and Airflow (part 1)”

Airflow: create and manage Data Pipelines easily

This bootstrap guide was originally published at GoSmarten but as the use cases continue to increase, it’s a good idea to share it here as well.

What is Airflow

The need to perform operations or tasks, either simple and isolated or complex and sequential, is present in all things data nowadays. If you or your team work with lots of data on a daily basis there is a good chance you’re struggled with the need to implement some sort of pipeline to structure these routines. To make this process more efficient, Airbnb developed an internal project conveniently called Airflow which was later fostered under the Apache Incubator program. In 2015 Airbnb open-sourced the code to the community and, albeit its trustworthy origin played a role in its popularity, there are many other reasons why it became widely adopted (in the engineering community). It allows for tasks to be set up purely in Python (or as Bash commands/scripts).

What you’ll find in this tutorial

Not only we will walk you through setting up Airflow locally, but you’ll do so using Docker, which will optimize the conditions to learn locally while minimizing transition efforts into production. Docker is a subject for itself and we could dedicate a few dozen posts to, however, it is also simple enough and has all the magic needed to show off some of its advantages combining it with Airflow. Continue reading “Airflow: create and manage Data Pipelines easily”

Using Akka Streaming for “saving alerts” – part 2

This blog post is the second and final part of the post Using akka streaming for “saving alerts”. Part 1 is available here.  In this part we enter the details on how the application was designed.

Full disclosure: this post was initially published at Bonial tech blog here. If you are looking for positions in tech, I would definitely recommend checking their career page.

Application Actor System

The following illustration gives you a schematic view of all the actors used in our application, and (hopefully) some of the mechanics of their interaction:


As previously mentioned, one could divide the application lifecycle logically into three main stages. We will get into more detail about each one of them next, but for now let us walk through them narrowly and try to map to our application Continue reading “Using Akka Streaming for “saving alerts” – part 2″

Getting Started with Spark (part 3) – UDFs & Window functions

This post attempts to continue the previous introductory series “Getting started with Spark in Python” with the topics UDFs and Window Functions.

Note: For this post I’m using Spark 1.6.1. There are some minor differences in comparison to the new coming Spark 2.0, such as using a SparkSession object to initialize the Spark Context, instead of HiveContext as I do here. Nonetheless, the important parts are common in both.

Continue reading “Getting Started with Spark (part 3) – UDFs & Window functions”

Spark 2.0: From quasi to proper-streaming?

This post attempts to follow the relatively recent new Spark release – Spark 2.0 – and understand the differences regarding Streaming applications. Why is streaming in particular?, you may ask. Well, Streaming is the ideal scenario most companies would like to use, and the competition landscape is definitely heating up, specially with Apache Flink and Google’s Apache Beam.

Why is streaming so difficult

There are three main problems when it comes to building real time applications based on streaming data:

  • Data consistency:  due to the distributed architecture nature it is possible that at any given point in time some events have been processed in some nodes and not  in other nodes, even though these events might actually have occurred before than others. For example, you may have a Logout event without a Login event, close app event without open app, etc.
  • Fault tolerance (FT): related to the first problem, on node failure processing engines may try to reprocess an event that had actually already been processed by the failed node, leading to duplicated records and/or inaccurate metrics
  • Dealing with late data/out-of-order data: either because you are reading from a message bus system such as Kafka or Kinesis, or simply because mobile app might only be able to sync data hours later, developers simply must tackle this challenge in application code.

See this post for an excellent detailed explanation. Continue reading “Spark 2.0: From quasi to proper-streaming?”

Overview HP Vertica vs AWS Redshift

While working in HP some years ago, I was exposed to not only internal training materials, but also a demo environment. I still remember the excitement when HP acquired Vertica Systems in 2011, and we had a new toy to play with… Come on, you can’t blame me, distributed DBs was something only the cool kids were doing.

Bottom line is that it’s been a while since I laid eyes on it… Well recently, while considering possible architectural solutions, I had the pleasure  to revisit Vertica. And since AWS Redshift has been gaining a lot of popularity and we’re also using it at some of our clients, I thought I might give some easy summary to help others.

Now if you’re expecting a smack down post, then I’m afraid I’ll disappoint you – for that you have the internet. If experience has taught me  something is that – in the case of top-notch solutions there are only use cases, and one finds the best fitting one. Continue reading “Overview HP Vertica vs AWS Redshift”

Spark – Redshift: AWS Roles to the rescue

If you are using AWS to host your applications, you probably heard that you can apply IAM Roles also to ec2 instances. In a lot of cases this can be a really cool way to avoid passing AWS credentials to your applications, and having the pain of having to manage key distribution among servers, as well as ensuring key rotation mechanisms for security purposes.

This post is about a simple trick on how to take advantage of this feature when your Spark job needs to interact with AWS Redshift.

As can be read in Databricks repo for Spark-redshift library the are three (3) strategies for setting up AWS credentials: either setup in hadoop configuration (how many people are used to so far with Cloudera or HortonWorks), encoding the keys in a tempdir (by far not the best option if you ask me), or using temporary keys. The last strategy is the one being discussed here, and its based on AWS own documentation how to use temporary keys. Continue reading “Spark – Redshift: AWS Roles to the rescue”

Distcp S3-hdfs for eu-central-1 (AWS Frankfurt) – details “behind the trenches”


Distcp (distributed copy) is a fairly old tool used to move a large quantity of files usually within hdfs, using MapReduce job to do so where mappers list the source files and reducers do the copy heavy lifting. Another useful integration is that it can also deal with file migrations between hdfs and AWS S3. In the meanwhile S3Distcp came along as an specific enhancement for speeding up these last kind of migrations.

However, what might be new for you is the requirement to use either one of these tools to move data from/to a bucket located in S3 Frankfurt region.   So this post intends to save you that half an hour of painful google research trying to figure out why your Oozie job is failing.

Without further ado, before you dive in to your Cloudera or Hortonworks Manager to edit hadoop properties in core-site.xml, I would suggest that you SSH to one of your hadop nodes, and start by testing running distcp command. Here is the sintax for moving data from S3 to HDFS:

hadoop distcp -Dfs.s3a.awsAccessKeyId=XXX \
-Dfs.s3a.awsSecretAccessKey=XX \ s3a://YOUR-BUCKET-NAME/ \

So yeah, in case you haven’t noticed, the black magic pieces in this command that makes things work just like in other regions are: the “fs.s3a.endpoint” property, as well as the “a” after “s3”.

Also in case you are moving data from hdfs to S3, besides write access, make sure you also give remove permissions on your AWS policy attaches to  the given user account on for that S3 bucket (or specific path) . This is because Distcp produces temporary files adjacent to your bucket target file migration, which you want to allow the reducers to remove in the end. Here is an example of an AWS policy for that effect:

    "Version": "2012-10-17",
    "Statement": [
            "Effect": "Allow",
            "Action": [
            "Resource": "arn:aws:s3:::*"
            "Effect": "Allow",
            "Action": [
            "Resource": [
                "arn:aws:s3:::YOUR-BUCKET" ]

Alternatively you might also want to test it using S3Distcp tool. Here is another example:

 hadoop jar s3distcp.jar --src /data/ \
--dest s3a://YOUR-BUCKET-NAME/ \

Note that the s3distcp jar needs to be locally on the host file system from which you are running the command. If you have the jar in hdfs, here is an example how you can fetch it:

hadoop fs -copyToLocal /YOUR-JARS-LOCATION-IN-HDFS/s3distcp.jar \

Also there is also the possibility you might stumble uppon some headaches with S3Distcp tool. This might prove itself to be handy.

Hope this helped! Where to go next? AWS has a handy best practice guide listing different strategies for data migration here.



Getting started with the Spark (part 2) – SparkSQL

Update: this tutorial has been updated mainly up to Spark 1.6.2 (with a minor detail regarding Spark 2.0), which is not the most recent version of Spark at the moment of updating of this post. Nonetheless, for the operations exemplified you can pretty much rest assured that the API has not changed substantially. I will try to do my best to update and point out these differences when they occur.

SparkSQL has become one of the core modules of Spark ecosystem (with its API already available for all currently supported languages, namely Scala, Java, Python, and R), and is intended to allow you to structure data into named columns (instead of with raw data without a schema as with RDDs) which can be queried with SQL syntax taking advantage of a distributed query engine running in the background.

In practical terms, you can very much think of it as equivalent as an attempt to emulate the API’s of popular dataframes in R or Python pandas, as well as SQL tables. However, if I do my job correctly, I succeed in proving you that they are not the same (and hopefully more).

In order for you to use SparkSQL you need to import SQLContext. Let’s start how you would do it with Python:

# import required libs
from pyspark import SparkConf, SparkContext
# here is the import that allows you to use dataframes
from pyspark.sql import SQLContext

# initialize context in Spark 1.6.2
conf = SparkConf().setMaster('local[2]').setAppName('SparkSQL-demo')
sc = SparkContext(conf=conf)
sqlContext = SQLContext(sc)

Note: in the new Spark 2.0, this would indeed vary as such:

# import required libs
from pyspark.sql import SparkSession
# initialize context in Spark 2.0
<span class="n">spark</span> <span class="o">=</span> <span class="n">SparkSession</span>\
    <span class="o">.</span><span class="n">builder</span>\
    <span class="o">.</span><span class="n">appName</span><span class="p">(</span>'SparkSQL-demo')<span class="o">.</span><span class="n">getOrCreate</span><span class="p">()</span>

Good news, Scala does not vary much:

//import required libs
import org.apache.spark.{SparkConf, SparkContext}
//here is the import that allows you to use dataframes
import org.apache.spark.sql.SQLContext

# initialize context in Spark 1.6.2
val conf = new SparkConf().setMaster(&quot;local[2]&quot;).setAppName(&quot;SparkSQL-demo&quot;)
val sc = new SparkContext(conf=conf)
val sqlContext = new SQLContext(sc)


A dataframe can be created from an existing RDD, as well as similarly to RDDs from a variety of different sources such astext files, HDFS, S3,  Hive tables, external databases, etc. Here’s an example how to build a dataframe by reading from S3 (in Python):

df ='s3n://my-bucket/example/file.json')


Note: we use “s3n” prefix which is valid for a serious of regions in AWS, but not valid for all of them. For example, to access S3 buckets in Frankfurt region you would rather use “s3a” as a prefix.

What sets Spark Dataframes apart

However, and even though the goal is to hide the maximum amount of disparities, there are nonetheless big conceptual differences between a Spark Dataframe and a Pandas dataframe. The first is that the immutability side has (thank God) not been removed. In other words, when you call an action on an Dataframe, you do need to save it into a variable. I’ve seen a lot of data scientists stumble on this in their first Spark days, so let us illustrate this in a very simple example:

# Create a Dataframe with pseudo data:
column_names = [&quot;customer_name&quot;, &quot;date&quot;, &quot;category&quot;,
&quot;product_name&quot;, &quot;quantity&quot;, &quot;price&quot;]
# Note that we create an RDD, and then cast it to
# Dataframe by providing a schema
column_names = ["customer_name", "date", "category",
"product_name", "quantity", "price"]
df01 = sc.parallelize(
[("Geoffrey", "2016-04-22", "A", "apples", 1, 50.00),
("Yann", "2016-04-22", "B", "Lamp", 1, 38.00)]).toDF(column_names)

df02 ="customer_name", "date", "category", "product_name", "price", (df01['quantity']+1).alias("quantity"))

|customer_name|      date|category|product_name|quantity|price|
|     Geoffrey|2016-04-22|       A|      apples|       1| 50.0|
|         Yann|2016-04-22|       B|        Lamp|       1| 38.0|

# .. where df02 is, as expected 

|customer_name|      date|category|product_name|price|quantity|
|     Geoffrey|2016-04-22|       A|      apples| 50.0|       2|
|         Yann|2016-04-22|       B|        Lamp| 38.0|       2|

The second most obvious difference (compared to pandas df) is the distributed nature of a Spark Dataframe, similarly as RDDs are. As a matter a fact,in previous versions of Spark Dataframes were even called SchemaRDD. So internally the same redundancy principles of RDDs apply, you should in fact still keep in mind the Driver – Executor programming model, as well as Actions and Transformations concept.

Also important to note is that Spark Dataframes use internally a query optmizer, one of the reasons why the performance difference between native Scala written Spark code vs Python or Java drops significantly as can be seen in the next illustration provided by databricks:




Let us view some useful operations one can perform with spark dfs. For the sake of brevity I will continue this tutorial mainly using the python API, as the syntax does not really vary that much.

Let us create a bigger fake (and complete utterly non-sense) dataset to make operations more interesting:

customers = sc.parallelize([("Geoffrey", "2016-04-22", "A", "apples", 1, 50.00),
("Geoffrey", "2016-05-03", "B", "Lamp", 2, 38.00),
("Geoffrey", "2016-05-03", "D", "Solar Pannel", 1, 29.00),
("Geoffrey", "2016-05-03", "A", "apples", 3, 50.00),
("Geoffrey", "2016-05-03", "C", "Rice", 5, 15.00),
("Geoffrey", "2016-06-05", "A", "apples", 5, 50.00),
("Geoffrey", "2016-06-05", "A", "bananas", 5, 55.00),
("Geoffrey", "2016-06-15", "Y", "Motor skate", 7, 68.00),
("Geoffrey", "2016-06-15", "E", "Book: The noose", 1, 125.00),
("Yann", "2016-04-22", "B", "Lamp", 1, 38.00),
("Yann", "2016-05-03", "Y", "Motor skate", 1, 68.00),
("Yann", "2016-05-03", "D", "Recycle bin", 5, 27.00),
("Yann", "2016-05-03", "C", "Rice", 15, 15.00),
("Yann", "2016-04-02", "A", "bananas", 3, 55.00),
("Yann", "2016-04-02", "B", "Lamp", 2, 38.00),
("Yann", "2016-04-03", "E", "Book: Crime and Punishment", 5, 100.00),
("Yann", "2016-04-13", "E", "Book: The noose", 5, 125.00),
("Yann", "2016-04-27", "D", "Solar Pannel", 5, 29.00),
("Yann", "2016-05-27", "D", "Recycle bin", 5, 27.00),
("Yann", "2016-05-27", "A", "bananas", 3, 55.00),
("Yann", "2016-05-01", "Y", "Motor skate", 1, 68.00),
("Yann", "2016-06-07", "Z", "space ship", 1, 227.00),
("Yoshua", "2016-02-07", "Z", "space ship", 2, 227.00),
("Yoshua", "2016-02-14", "A", "bananas", 9, 55.00),
("Yoshua", "2016-02-14", "B", "Lamp", 2, 38.00),
("Yoshua", "2016-02-14", "A", "apples", 10, 55.00),
("Yoshua", "2016-03-07", "Z", "space ship", 5, 227.00),
("Yoshua", "2016-04-07", "Y", "Motor skate", 4, 68.00),
("Yoshua", "2016-04-07", "D", "Recycle bin", 5, 27.00),
("Yoshua", "2016-04-07", "C", "Rice", 5, 15.00),
("Yoshua", "2016-04-07", "A", "bananas", 9, 55.00),
("Jurgen", "2016-05-01", "Z", "space ship", 1, 227.00),
("Jurgen", "2016-05-01", "A", "bananas", 5, 55.00),
("Jurgen", "2016-05-08", "A", "bananas", 5, 55.00),
("Jurgen", "2016-05-08", "Y", "Motor skate", 1, 68.00),
("Jurgen", "2016-06-05", "A", "bananas", 5, 55.00),
("Jurgen", "2016-06-05", "C", "Rice", 5, 15.00),
("Jurgen", "2016-06-05", "Y", "Motor skate", 2, 68.00),
("Jurgen", "2016-06-05", "D", "Recycle bin", 5, 27.00),
]).toDF(["customer_name", "date", "category", "product_name", "quantity", "price"])


The dataset is minimal for the example, so it really does not make a difference to cache it in memory. But my point is to illustrate that similarly to RDDs, you should also care about memory persistence, and can indeed use a similar API. Moreover, since next we are going to use several distinct operations of type “Action” (yielding back results to the driver), it is of good practice to signal that this dataframe should be kept in memory for further operations.

# False
# True

OK, so yes, there are some exceptions, and in this case as you can see the cache() method does not obey the laws of immutability. Moving on, let’s take a peak on it:

|customer_name|      date|category|   product_name|quantity|price|
|     Geoffrey|2016-04-22|       A|         apples|       1| 50.0|
|     Geoffrey|2016-05-03|       B|           Lamp|       2| 38.0|
|     Geoffrey|2016-05-03|       D|   Solar Pannel|       1| 29.0|
|     Geoffrey|2016-05-03|       A|         apples|       3| 50.0|
|     Geoffrey|2016-05-03|       C|           Rice|       5| 15.0|
|     Geoffrey|2016-06-05|       A|         apples|       5| 50.0|
|     Geoffrey|2016-06-05|       A|        bananas|       5| 55.0|
|     Geoffrey|2016-06-15|       Y|    Motor skate|       7| 68.0|
|     Geoffrey|2016-06-15|       E|Book: The noose|       1|125.0|
|         Yann|2016-04-22|       B|           Lamp|       1| 38.0|
only showing top 10 rows


As previously referred, one advantage over RDDs is the ability to use standard SQL syntax to slice and dice over a Spark Dataframe. Let’s check for example all distinct products that we have:

spark_sql_table = 'customers'
sql_qry = "select distinct(product_name) from {table}".format(table=spark_sql_table)
distinct_product_names_df01 = sqlContext.sql(sql_qry)

|product_name              |
|space ship                |
|Book: The noose           |
|Rice                      |
|Motor skate               |
|Book: Crime and Punishment|
|Recycle bin               |
|Lamp                      |
|bananas                   |
|Solar Pannel              |
|apples                    |


Similarly to how we work with RDDs, with Spark DataFrames you are not restricted to SQL, and can indeed call methods on dataframes, which yields a new dataframe (remember the immutability story, right?)

Let’s check for example the same previous example on how to get all distinct:

distinct_product_names_df02 ='product_name').distinct()

|product_name |
|space ship |
|Book: The noose |
|Rice |
|Motor skate |
|Book: Crime and Punishment|
|Recycle bin |
|Lamp |
|bananas |
|Solar Pannel |
|apples |


Note that you can as well work on a dataframe just like you iterate on an RDD, which causes the dataframe to be cast into a RDD in the background:

distinct_product_names_rdd = row: row[3])
# calling will yield an exception
except Exception as err:
 print('Error: {}'.format(err))
 print('BUT calling .take() this works, as it is an RDD now: {}' \
 .format(distinct_product_names_rdd.take(3)) )

distinct_product_names = distinct_product_names_rdd.distinct().collect()
print('Distinct product names are: {}'.format(distinct_product_names))
print('\n... And, as expected, they are {} in total.' \

Error: 'PipelinedRDD' object has no attribute 'show'
BUT calling .take() this works, as it is an RDD now: ['apples', 'Lamp', 'Solar Pannel']
Distinct product names are: ['Rice', 'bananas', 'space ship', 'Book: The noose', 'Solar Pannel', 'Recycle bin', 'apples', 'Lamp', 'Book: Crime and Punishment', 'Motor skate']

... And, as expected, they are 10 in total.


Similar than in R or Python, you can have summary statistics. Let’s get statistics for Category, quantity and price columns:

customers.describe('category', 'quantity', 'price').show()
|summary|category|          quantity|             price|
|  count|      39|                39|                39|
|   mean|    null| 4.153846153846154| 68.94871794871794|
| stddev|    null|2.9694803863945936|59.759583559965165|
|    min|       A|                 1|              15.0|
|    max|       Z|                15|             227.0|


Being a categorical variable, as you can see the summary regarding the ‘category’ column does not make much sense. We’ll work on better on this later, but for now, let’s do something simple and try to look into product_name and category.

Let’s suppose we are interested in finding out how does purchasing behavior vary among categories, namely what is the average frequency of purchases per category.

cat_freq_per_cust = customers.stat.crosstab('customer_name', 'category')

|customer_name_category|  Y|  Z|  A|  B|  C|  D|  E|
|              Geoffrey|  1|  0|  4|  1|  1|  1|  1|
|                Yoshua|  1|  2|  3|  1|  1|  1|  0|
|                  Yann|  2|  1|  2|  2|  1|  3|  2|
|                Jurgen|  2|  1|  3|  0|  1|  1|  0|


cols = cat_freq_per_cust.columns

|summary|Y                 |Z                |A                |B                |C  |D                 |E                 |
|count  |4                 |4                |4                |4                |4  |4                 |4                 |
|mean   |1.5               |1.0              |3.0              |1.0              |1.0|1.5               |0.75              |
|stddev |0.5773502691896257|0.816496580927726|0.816496580927726|0.816496580927726|0.0|0.9999999999999999|0.9574271077563381|
|min    |1                 |0                |2                |0                |1  |1                 |0                 |
|max    |2                 |2                |4                |2                |1  |3                 |2                 |


Note: “count” column does not have much meaning here; it simply counts number of rows blindly, and as expected since the rows without a given value are filled with “zero” instead of Null, it counts them too.

Let’s check the same for products.

# Let's view products vertically, for formatting reasons
customers.stat.crosstab('product_name', 'customer_name' )\
|Book: Crime and Punishment|0       |0     |1   |0     |
|Recycle bin               |0       |1     |2   |1     |
|Solar Pannel              |1       |0     |1   |0     |
|space ship                |0       |1     |1   |2     |
|Motor skate               |1       |2     |2   |1     |
|Rice                      |1       |1     |1   |1     |
|apples                    |3       |0     |0   |1     |
|Lamp                      |1       |0     |2   |1     |
|bananas                   |1       |3     |2   |2     |
|Book: The noose           |1       |0     |1   |0     |


freq_products = customers.stat.crosstab('customer_name', 'product_name')
freq_prod_cols = freq_products.columns

|summary|          bananas|       space ship|   Book: The noose|Rice|       Motor skate|             Lamp|            apples|Book: Crime and Punishment|      Recycle bin|      Solar Pannel|
|  count|                4|                4|                 4|   4|                 4|                4|                 4|                         4|                4|                 4|
|   mean|              2.0|              1.0|               0.5| 1.0|               1.5|              1.0|               1.0|                      0.25|              1.0|               0.5|
| stddev|0.816496580927726|0.816496580927726|0.5773502691896257| 0.0|0.5773502691896257|0.816496580927726|1.4142135623730951|                       0.5|0.816496580927726|0.5773502691896257|
|    min|                1|                0|                 0|   1|                 1|                0|                 0|                         0|                0|                 0|
|    max|                3|                2|                 1|   1|                 2|                2|                 3|                         1|                2|                 1|

Alternatively we could have answered suing a more SQL-like approach:

cust_prod = customers.groupBy('customer_name', 'product_name') \
 .agg(func.count('product_name')) \
 .orderBy('product_name', 'customer_name'), truncate=False)

|customer_name|product_name              |count(product_name)|
|Yann         |Book: Crime and Punishment|1                  |
|Geoffrey     |Book: The noose           |1                  |
|Yann         |Book: The noose           |1                  |
|Geoffrey     |Lamp                      |1                  |
|Yann         |Lamp                      |2                  |
|Yoshua       |Lamp                      |1                  |
|Geoffrey     |Motor skate               |1                  |
|Jurgen       |Motor skate               |2                  |
|Yann         |Motor skate               |2                  |
|Yoshua       |Motor skate               |1                  |
|Jurgen       |Recycle bin               |1                  |
|Yann         |Recycle bin               |2                  |
|Yoshua       |Recycle bin               |1                  |
|Geoffrey     |Rice                      |1                  |
|Jurgen       |Rice                      |1                  |
|Yann         |Rice                      |1                  |
|Yoshua       |Rice                      |1                  |
|Geoffrey     |Solar Pannel              |1                  |
|Yann         |Solar Pannel              |1                  |
|Geoffrey     |apples                    |3                  |
|Yoshua       |apples                    |1                  |
|Geoffrey     |bananas                   |1                  |
|Jurgen       |bananas                   |3                  |
|Yann         |bananas                   |2                  |
|Yoshua       |bananas                   |2                  |
|Jurgen       |space ship                |1                  |
|Yann         |space ship                |1                  |
|Yoshua       |space ship                |2                  |


cust_prod.groupBy('product_name') \
 .agg(func.avg('count(product_name)')) \
|product_name              |avg(count(product_name))|
|space ship                |1.3333333333333333      |
|Book: The noose           |1.0                     |
|Rice                      |1.0                     |
|Motor skate               |1.5                     |
|Book: Crime and Punishment|1.0                     |
|Recycle bin               |1.3333333333333333      |
|Lamp                      |1.3333333333333333      |
|bananas                   |2.0                     |
|Solar Pannel              |1.0                     |
|apples                    |2.0                     |

Note that this does not take into account the quantity of each product bought, just how many times client was there to buy a given item. For our last question that may be OK, because we understand the different nature of products.
But let us assume it is not OK, and we do need to count by quantity. For that we can use pivot dataframes, a feature introduced in Spark 1.6

total_products_bought_by_customer = customers.groupBy('customer_name')\

|customer_name|Book: Crime and Punishment|Book: The noose|Lamp|Motor skate|Recycle bin|Rice|Solar Pannel|apples|bananas|space ship|
|         Yann|                         5|              5|   3|          2|         10|  15|           5|  null|      6|         1|
|       Yoshua|                      null|           null|   2|          4|          5|   5|        null|    10|     18|         7|
|     Geoffrey|                      null|              1|   2|          7|       null|   5|           1|     9|      5|      null|
|       Jurgen|                      null|           null|null|          3|          5|   5|        null|  null|     15|         1|


Hope this has been a good introduction, and I suggest that you further check out this great post from databricks covering similar operations.

For the last part of this tutorial series I plan to cover window functions, something that until the more recent Spark 2.0 was possible thanks to HiveContext.