A stream processing exercise.
Kinesis can be used as an Event Collector Service.
The clients, for example mobile devices, push events directly into a Kinesis Stream.
In this exercise the incoming events are generated from a file or randomly.
Start with an implementation as simple as possible.
For the consumer use a small python script which reads from a downloaded file and writes line-by-line after being converted to an event.
Write the data into a Kinesis stream.
The consumer would be another simple python script.
Pull the events from the Kinesis stream.
Tag all events seen within a single minute with the same key value.
Next steps are a groupUntil with a defined timeout and the write of result to S3.
In case events should be counted by type and minute a composite key can be used. In Python use a tuple ((“usa”, “201608211700”), 1).
The events per minute per worker would be on S3 in real-time.
Simplest design consists of two modules.
The producer, which reads the data from the file sequentially.
Transforms TSV raw data into an event.
Creates a batch of event data.
Then pushes a batch into the Kinesis stream.
A consumer, which polls from the Kinesis stream.
Aggregates the event data.
And writes the result into persistent storage.
use virtualenvwrapper and create a virtual environment for python.
mkvirtualenv streams
install the packages into the virtual environment
pip install -r requirements.txt
configure the AWS environment locally.
aws configure
create a Kinesis stream in AWS.
aws kinesis create-stream --stream-name <your-stream> --shard-count 1
run the producer and consumer in different processes connected by a queue
configure the consumer and producer via inject in run_stream_processing.py.
python run_stream_processing.py
producer can either generate infinite stream of random events or finite stream of events from a file.
it is possible to write the events to console, local queue or Kinesis stream.
change inject.configure() in producer.py to activate the suited configuration.
python producer.py
to test differently keyed (timestamp, random) events change inject.configure() in consumer.py.
running the consumer this way only makes sense if the process can connect to the configured stream, for example a Kinesis Stream.
python consumer.py