josephmachado/data_engineering_best_practices_log

Code to demonstrate data engineering metadata & logging best practices

Data Engineering Best Practices - #2. Metadata & Logging

Code for blog at Data Engineering Best Practices - #2. Metadata & Logging

This is part of a series of posts about data engineering best practices:

1. Data Engineering Best Practices - #1. Data flow & Code
2. Data Engineering Best Practices - #2. Metadata & Logging

Project

For project overview and architecture refer to this Data flow & code repo.

Logging architecture

Setup

If you'd like to code along, you'll need

Prerequisite:

1. git version >= 2.37.1
2. Docker version >= 20.10.17 and Docker compose v2 version >= v2.10.2. Make sure that docker is running using docker ps
3. pgcli

Run the following commands via the terminal. If you are using Windows, use WSL to set up Ubuntu and run the following commands via that terminal.

git clone https://github.com/josephmachado/data_engineering_best_practices_log.git
cd data_engineering_best_practices_log
make up # Spin up containers
make ddl # Create tables & views
make ci # Run checks & tests
make etl # Run etl
make spark-sh # Spark shell to check created tables

spark.sql("select partition from adventureworks.sales_mart group by 1").show() // should be the number of times you ran `make etl`
spark.sql("select count(*) from businessintelligence.sales_mart").show() // 59
spark.sql("select count(*) from adventureworks.dim_customer").show() // 1000 * num of etl runs
spark.sql("select count(*) from adventureworks.fct_orders").show() // 10000 * num of etl runs
:q // Quit scala shell

You can see the results of DQ checks and metadata as shown below. Open the metadata cli using make meta

select * from ge_validations_store limit 1;
select * from run_metadata limit 2;
exit

Use make down to spin down containers.

Infrastructure components

1. Spark applications: We have spark standalone cluster. When we submit a spark job a new spark application will be created and its UI will be available at localhost:4040
2. Metadata DB: We have a postgres container that is used to store results of data quality checks (run by Great Expectations) and we store run information (table: run_metadata) in this database as well. You can access the metadata db using the make meta command.
3. Prometheus: We have a Prometheus server running, and we have a Prometheus job that runs ever 5s (configured here) to pull spark metrics (via Spark PrometheusServlet). Prometheus is available at localhost:9090.
4. Grafana: We have a Grafana service running as the UI for prometheus data. Grafana is available at localhost:3000, with username admin and password spark.

Future Work

1. Setup Dashboard configuration for Grafana to display Spark metrics.
2. Move log storage from local filesystem to a service like Grafana Loki.
3. Display metadata and Data quality results in Grafana UI.
4. Add type information or make the metadata into a JSON.