Skip to content

Latest commit

 

History

History
66 lines (48 loc) · 1.97 KB

README.md

File metadata and controls

66 lines (48 loc) · 1.97 KB

apache-sample

Apache data system written for development in a local cluster and incramental deployment modifications for production environments. Great for learning about cloud-native development in the apache software ecosystem.

System Design

  • Kafka event source to ingest realtime application data
  • Spark framework for microbatch and batch processes
  • Delta ACID-compliant storage layer on file storage
  • Hive metadata store for the delta schemas
  • Trino analytics query engine for ad-hoc analysis

System Tests

An end-to-end test of the system can be run in kubernetes. The test:

  1. publishes sample data to a topic on the kafka cluster
  2. ingests the data into a staging table on delta using structured-streaming
  3. performs windowed aggregations on the data and saves the results
  4. triggers a sql analytics query through trino to simulate an analyst

The tests are triggered through github actions, although you will need to use a self-hosted runner.

Developer Notes

Testing in the Development Environment

Ensure docker desktop is running

open -a Docker;
kubectl config use-context docker-desktop

Build docker images for the kafka, spark, and trino jobs

make kafka-producer-image;
make spark-jobs-image;
make trino-queries-image

Run the system test

cd scripts;
./01-install-operators.sh;
./02-deploy-kafka.sh;
./03-deploy-delta-trino.sh;
./04-run-spark-jobs.sh;
./05-run-trino-query.sh

Modifying for a Production Environment

  1. Scale the kafka cluster and spark streaming jobs
  2. Change the batch job deployment to run on an ephemeral cluster
  3. Change the storage configuration to cloud storage
  4. Update the hive metastore to postgresql
  5. Scale the trino cluster for more performant queries
  6. Setup an orchestrator such as Airflow or Prefect
  7. Configure security and monitoring for the application

Updates

Checkout this open-source project and company: stackable.