20 March 2021
2 minutes to read
I’ve gotten into reading most articles I had for the past weeks. I’m seeing more and more regarding streaming pipelines although I think there’s much to be done in batch. In the next weeks I’ll be taking a closer look into DBT and it’s internal, to better understand the best way to integrate it into my current work. The tool is great to work in a single data warehouse but isn’t ready yet for a data mesh.
Cleaning Up Your Postgres Database gives really good tips for postgreSQL. In recent weeks I’ve tried to move millions of events into postgres and can atest that this isn’t a really great use case for postgres 😅.
AWS also gives some tips on how to store timeseries tables on Postgres on Designing high-performance time series data tables on Amazon RDS for PostgreSQL.
TIL: B-tree index is great but can increase in size linearly. BRIN index tracks the minimum and maximum time values over a range so it can be a great match for time-series databases (in doubt use B-tree)!
For OLTP databases, Postgres and mysql are some of the best known (open source is a plus). After reading Performance differences between Postgres and MySQL I’ve actually gotten some good insights into their internals.
It’s very easy to develop some dashboards but it can be a bit harder to make it useful. In Best practices for BI dashboards, Metabase gives some tips on how to develop a dashboard that actually helps in the decision making process.
Spotify gives an overview on their experimentation framework on Spotify’s New Experimentation Coordination Strategy. This is a long way from simple A/B testing and makes the case for deeper analytics to improve a product.
Flink gives a nice overview on Batch Execution Mode using their unified API.
TIL: Difference of tumbling window and a hopping window in a streaming pipeline 😎
Airflow is a great tool but, to be correctly used, needs to thought as a worflow tool and not as a data processor. The article TaskFlow API in Apache Airflow 2.0 — Should You Use It? starts by explaining what is the taskflow API but ends with a warning. Airflow, specially in 2.0, is great to manage multiple processes, but we are best saving state and doing heavy computations with other tools like spark and Presto.
Stay safe, stay well 🙌
I'm José Cabeda, a data engineer focused on improving data systems and educating on how to use them. I also do a lot of planning and read as much as I can.