DEV Community

Francesco Tisiot
Francesco Tisiot

Posted on • Originally published at aiven.io on

Create your own data stream for Kafka with Python and Faker

image

We've all been there: we installed or purchased a new and shiny data management tool in order to test it, but immediately realised we don't have any data to use for our trial! Maybe we know the data's schema but we can't use our company's datasets for a number of reasons.

How can we give the platform a proper try? What if I tell you all you need are a few lines of Python code?
We'll use the real-world case of a pizza delivery chain to demonstrate how easy it is to produce proper fake data.

TL;DR: If you're just interested in a ready-to-go pizza-based Kafka Producer, check out the related Github repo!

(Quick links here:

Kafka setup

Python client settings

Enjoy!)

Data platforms are empty shells

Databases, data lakes, and datastores in general are cool pieces of technology allowing anyone to manage and analyze data, but they share a common problem: they are completely empty by default. This makes them hard to use for testing, which is still needed to evaluate whether a solution fits your use case.
Several platforms solve this problem by offering pre-filled sample datasets that can be installed or enabled with the help of few commands. Those datasets, however, are pretty static and do not always provide the set of features, fields or cardinality you need to perform your trials.

With Apache Kafka, the scale of this problem is even larger. Not only is it empty by default, but it is also a streaming data platform that works by ingesting, transforming and distributing data on the fly - and it expects a continuous flow of data. Finding streaming data sources is a hard task, and, especially if you just want to test the basics of the platform, setting them up properly can be quite cumbersome.

Creating fake data by hand is also not trivial. Even if you know your data schema, creating a coherent set of rows from scratch is challenging. In this blog post I'll walk you through how to create such a fake dataset for Kafka with a hot topic in mind: Pizza!

Kafka setup

Let's start with the tech setup. For this example we'll need a Kafka cluster. Creating it on Aiven.io is really easy:

  1. Navigate to Aiven.io console and sign in (or sign up; it's quick and easy).
  2. Click + Create a new service
  3. Select the Kafka service (if we're picky we can also choose our favorite Apache Kafka version)
  4. Select the Cloud provider we want to deploy our services to, together with the Cloud region
  5. Select the service plan based on our needs
  6. Give the service a name

Since I'm based in Italy I could easily go for the recently created aws-eu-south AWS region, located in Milan, to minimize latency. Your choice depends on where you're located or where you plan to provide your services. For our initial test we're ok in using a Startup plan, knowing we can always upgrade (or downgrade) in the future.

Then we just click on Create Service. It will take a few minutes before our Kafka 3-node cluster shows up in the RUNNING state.

Getting the Service Credentials

While we're waiting for the service to be up and running, we can already start preparing for the next step: downloading the certificates required to connect.

credentials

We can go to the Overview tab of our Kafka instance in Aiven.io console where we can easily find the Access Key, Access Certificate and CA Certificate. Then download them to our computer creates 3 files:

  • service.key: the Access Key
  • service.cert: the Access Certificate
  • ca.pem: the CA Certificate

On the Overview tab, I can also take note of the Service URI (usually in the form <INSTANCE_NAME>-<PROJECT_NAME>.aivencloud.come:<PORT>) that we'll use to correctly point our producer to the Kafka cluster.

Auto Topic Creation & Enabling Kafka REST APIs

By default, Kafka producers can push data only to pre-created topics. In order to allow topics to be created on the fly while pushing the first record, in this tutorial we'll enable the kafka.auto_create_topics_enable parameter in Aiven.io console Overview tab; scroll down to the Advanced configuration section and then click the + Add configuration option menu.

advanced_configurations

The last step needed in the configuration is to enable the Kafka REST API (Karapace) in Kafka's console Overview tab, for our cluster. This step is, strictly speaking, not fundamental, but will allow us to check our producer by reviewing the pushed records in Aiven.io console Topics tab.

Python client setting

We'll use the kafka-python client to build our producer. All we need to do is install it:

pip install kafka-python
Enter fullscreen mode Exit fullscreen mode

And then set a Producer. Add this code to a new main.py file

import json
from kafka import KafkaProducer

folderName = "~/kafkaCerts/kafka-pizza/"
producer = KafkaProducer(
    bootstrap_servers="<INSTANCE_NAME>-<PROJECT_NAME>.aivencloud.com:<PORT>",
    security_protocol="SSL",
    ssl_cafile=folderName+"ca.pem",
    ssl_certfile=folderName+"service.cert",
    ssl_keyfile=folderName+"service.key",
    value_serializer=lambda v: json.dumps(v).encode('ascii'),
    key_serializer=lambda v: json.dumps(v).encode('ascii')

)
Enter fullscreen mode Exit fullscreen mode

In the sample code we imported the dependencies and set the correct parameters like bootstrap_servers, ssl_cafile, ssl_certfile and ssl_keyfile which refer to the connection URI and the three certificate files mentioned in the section above.

The value_serializer and key_serializer parameters need a separate explanation. Later on, we will produce each record and key in Json format. To push it properly to Kafka we need to transform them to string format and encode. This is exactly what the code lambda v: json.dumps(v).encode('ascii') does.

We are now ready to push our first message to Kafka's test-topic with

producer.send("test-topic",
                key={"key": 1},
                value={"message": "hello world"}
            )
producer.flush()
Enter fullscreen mode Exit fullscreen mode

The flush() command blocks the code from executing until all async messages are sent.

If we did our homework correctly, we can now execute the main.py code with

python main.py
Enter fullscreen mode Exit fullscreen mode

We should be able to see in Aiven.io console, under the Topics tab, the topic correctly being created

test-topic

And after clicking on the topic name, on Messages, and selecting json as FORMAT we should be able to view our message

test-topic

Create Fake Datasets with Faker

So, let's back to our main topic: pizza.

We are the owners of a pizza delivery chain, and of course we want to push our orders to Apache Kafka. We know we receive calls, and note down the client's Name, Address and Phone Number (you never know, we might get lost while delivering). How can we mimic that information?

Welcome to Faker, a Python library allowing us to create proper fake data! We need to be on Python 3.5 and above, and we can install it with

pip install Faker
Enter fullscreen mode Exit fullscreen mode

We just need a simple code to create one (or more) tuple of data containing the Name, Address and Phone Number

from faker import Faker
fake = Faker()
message= {
        'name':fake.name(),
        'address':fake.address(),
        'phone':fake.phone_number()
        }
print(message)
Enter fullscreen mode Exit fullscreen mode

Which will print a record like the below

{
  'name': 'Adrian Cole', 
  'address': '9395 Smith Canyon\nSullivanport, UT 22200', 
  'phone': '001-959-108-3651'
}
Enter fullscreen mode Exit fullscreen mode

We can even localise the output by passing the locale as argument like

fake = Faker('it_IT')
Enter fullscreen mode Exit fullscreen mode

The same example above, localised in Italian with the it_IT parameter will generate

{
  'name': 'Sig. Leopoldo Piacentini', 
  'address': 'Piazza Cocci 707 Piano 3\nSesto Isabella lido, 53704 Forlì-Cesena (FE)', 
  'phone': '+39 12 26548428'
}
Enter fullscreen mode Exit fullscreen mode

Perfect, now our basic fake data generator is ready!

Well... a <Name, Address, Phone Number> tuple is not really rocket science, and also doesn't tell us anything about our business. We are a pizzeria, where are the pizzas?
Surprisingly Faker's standard providers do not include a pizza generator, but don't worry, we can create our own.

Create Custom Data Provider

We know we have a standard pizza menu consisting of few options, ranging from the traditional Margherita to the Mari & Monti mixing seafood and ham. Creating a fake pizza generator is just a matter of returning a random choice between the available options. We can create a new Faker provider in a separate pizzaproducer.py file.

import random
from faker.providers import BaseProvider

class PizzaProvider(BaseProvider):
    def pizza_name(self):
        validPizzaNames= ['Margherita',
                          'Marinara',
                          'Diavola',
                          'Mari & Monti',
                          'Salami',
                          'Pepperoni'
                        ]
        return validPizzaNames[random.randint(0,len(validPizzaNames)-1)]
Enter fullscreen mode Exit fullscreen mode

We can now import the PizzaProvider in our main.py file and run it for 10 samples

from pizzaproducer import PizzaProvider
fake.add_provider(PizzaProvider)
for i in range(0,10):
    print(fake.pizza_name())
Enter fullscreen mode Exit fullscreen mode

We correctly obtain

Mari & Monti
Salami
Marinara
Pepperoni
Marinara
Pepperoni
Salami
Pepperoni
Margherita
Pepperoni
Enter fullscreen mode Exit fullscreen mode

But like in any respectable pizzeria, we allow people to add toppings from a list, and similarly to the above, we can define a custom pizza_toppings function. The same goes to record which pizza shop in our chain is receiving the order, which generation is demanded to the pizza_shop function.

The full pizza fake data provider code can be found here, ready for a copy-paste into our pizzaproducer.py file.

Creating an order

We now have all the building blocks, let's create an order.
For each call, we note down the Name, Address and Phone Number. The customer however can order 1-10 pizzas, and, for each pizza, 0-5 additional toppings. To generate fake orders, can define a function that accepts a randomly generate orderid and returns the order message and related key.
One thing to note: we decided to key our messages with a representation of the shop name in Json format. This will ensure all orders from the same shop to appear in the same pizza-order topic partition, thus making sure that a shop's requests will be executed following the order arrival time.

# creating function to generate the pizza Order
def produce_pizza_order (orderid = 1):
    shop = fake.pizza_shop()
    # Each Order can have 1-10 pizzas in it
    pizzas = []
    for pizza in range(random.randint(1, MAX_NUMBER_PIZZAS_IN_ORDER)):
        # Each Pizza can have 0-5 additional toppings on it
        toppings = []
        for topping in range(random.randint(0, MAX_ADDITIONAL_TOPPINGS_IN_PIZZA)):
            toppings.append(fake.pizza_topping())
        pizzas.append({
            'pizzaName': fake.pizza_name(),
            'additionalToppings': toppings
        })
    # message composition
    message = {
        'id': orderid,
        'shop': shop,
        'name': fake.unique.name(),
        'phoneNumber': fake.unique.phone_number(),
        'address': fake.address(),
        'pizzas': pizzas
    }
    return message, key

Enter fullscreen mode Exit fullscreen mode

Calling the above code with

produce_pizza_order(704)
Enter fullscreen mode Exit fullscreen mode

Will generate a key like

{
  "shop":"Luigis Pizza"
}
Enter fullscreen mode Exit fullscreen mode

And a message like

{
  "id": 704,
  "shop": "Luigis Pizza",
  "name": "Jessica Green",
  "phoneNumber": "(549)966-3806x9591",
  "address": "458 Conway Dale Apt. 510\nZacharyborough, TX 48185",
  "pizzas": [
    {
      "pizzaName": "Mari & Monti",
      "additionalToppings": [
        "banana"
      ]
    },
    {
      "pizzaName": "Peperoni",
      "additionalToppings": [
        "ham"
      ]
    }
  ]
}

Enter fullscreen mode Exit fullscreen mode

Creating a Fake Producer

We described above two lego blocks: Kafka producer settings and fake order generator. What's missing?
We need a continuous stream of events. This we can easily simulate, for 100 messages for example, with a loop:

import time
while i <  100:
    message, key = produce_pizza_order(i)

    print("Sending: {}".format(message))
    # sending the message to Kafka
    producer.send(topic_name,
                  key=key,
                  value=message)
    # 2 seconds of sleep time before the next message
    time.sleep(2)

    # Force sending of all messages
    if (i % 100) == 0:
        producer.flush()
    i=i+1
producer.flush()
Enter fullscreen mode Exit fullscreen mode

The Code is Yours!

By following the above steps you should be able to correctly produce events to Kafka. But, if you are keen on looking at the ready-made project, check out the related Github repo which will enable you to generate data within minutes.

After executing the code, we can verify that the Producer did its job by going to the Topics tab in Aiven.io console and checking the pizza-orders offset:

offset

And, since we enabled Kafka REST APIs, also by viewing the actual topic content:

topic content

We love working with Kafka and our customers do too! if you want to know more, then check our resources:

Top comments (0)