Apache NiFi can be used to accelerate big data projects by allowing easy integration between various data sources. Using Apache NiFi it is easy to track what happened to your data (data provenance) and to provide features like guaranteed ordered delivery and error handling. In this example I’m going to configure NiFi to read COVID data from a REST API, split the data into individual records per country and publish the result to a Kafka topic. I’ve used the environment described here.
Calling a REST service
The InvokeHTTP processor can be used to call REST APIs. I’ve configured this as follows:
I’ve used a GET request on the following API: https://api.covid19api.com/summary. This gives you a JSON like:
Splitting the JSON
The SplitJson processor can split the input JSON into separate FlowFiles per country. This comes in handy when writing individual records to the Kafka topic.
Publishing records to Kafka
Below is the configuration for the PublishKafkaRecord processor. Not in the screenshot is Message Key Field. This indicates which field in the message is used as the key. I’ve used the Country field (“Country” as value for this field). Why did I use the country field? Messages with the same key are published to the same partition within Kafka and the ordering on a partition is guaranteed. When I consume messages, I consume from a specific partition. I can be sure I’ll consume them in the order they arrived per country.
Important to notice is that I use the JsonTreeReader in order to collect the entire JSON record. The JsonTreeReader and JsonRecordSetWriter are controller services which have a default configuration.
Result
When you enable your processors, you can run the processor which calls the REST API once. Of course, you can also schedule the execution of the InvokeHTTP processor to for example fetch data daily. It will read the data, the data will be split into individual FlowFiles and these FlowFiles will be produced to a Kafka topic which uses the country name as its key.
You can browse your Kafka topic with for example OffsetExplorer in order to check if the messages have arrived successfully.
You can also of course use the data provenance facilities by right clicking on a processor, “view data provenance”. This allows you to examine the input and output of every processor.
Finally
This simple example shows what you can achieve with relatively little effort using Apache NiFi. This is just the tip of the iceberg of this powerful product. Important to keep in mind when starting out with Apache NiFi is to look for simple solutions and not go overboard with many components, custom scripts, Jolt transformations, custom processors, etc when you do not really need them. This allows you and others to better understand what is going and also often makes the solution more efficient.
