Data Engineering Weekly #42

Weekly Data Engineering Newsletter

Welcome to the 42nd edition of the data engineering newsletter. This week's release is a new set of articles that focus on Benn's analytics is a mess, Earnest Research's experience with SQLFluff, eBay's new optimized Spark SQL Engine, Dropbox's optimizing payment with Machine Learning, Microsoft's how seasonality impact metrics, Lyft's end-to-end ML platform, Razorpay's Druid infrastructure, Astronomer's Airflow & Ray integration, Anna's take on how Slack shared channel can improve data quality, and confluent's Kafka summit recap.


Benn Stancil: Analytics is a mess

When we look at companies with mature data practices, we only see the final, stable metrics and dashboards. However, simple metrics like "What is the unique user count for this week"? the definition of unique can have multiple answers, and make no mistake, they all more or less correct. Are metrics real? Are we creating an analytical mess with multiple definitions of metrics? The author narrates how it's not only normal, but it's also necessary.

https://benn.substack.com/p/analytics-is-a-mess


Earnest Research: SQLFluff — The Linter For Modern SQL

SQL deserves linter more than ever, and I 100% agree. In this blog post, Earnest Research talks about its experience and effectiveness of SQLFluff, an open-source linter tool for SQL.

https://towardsdatascience.com/sqlfluff-the-linter-for-modern-sql-8f89bd2e9117

SQLFluff Github: https://github.com/sqlfluff/sqlfluff


eBay: Explore eBay’s New Optimized Spark SQL Engine for Interactive Analysis

eBay writes about its optimized SQL engine for interactive analysis. eBay effectively using the spark's thrift server on Yarn with workload isolation using Yarn queue. The usage of bloom filter indexing, transparent data caching strategy, bucketing improvements, and parquet read optimization are some of the exiting read.

https://tech.ebayinc.com/engineering/explore-ebays-new-optimized-spark-sql-engine-for-interactive-analysis/


Dropbox: Optimizing payments with machine learning

One of the challenges of the subscription business model is to manage the subscription renewal process efficiently to reduce involuntary churn. Dropbox writes an exciting case study of how it applied ML techniques in the renewal process to increase the retention rate.

https://dropbox.tech/machine-learning/optimizing-payments-with-machine-learning


Microsoft: How weekends can impact seasonality and metrics

The seasonality such as weekends, holidays are the critical factors to accommodate in the exploratory data analytics before interpreting the analysis. The blog narrates the walk through the impact of seasonality in analysis and discusses how to handle it. It might be overcomplicated or not fully necessary to get a formula to “normalize” such data. However, it might be helpful to track such seasonality to understand better how your business is doing.

https://medium.com/data-science-at-microsoft/how-weekends-can-impact-seasonality-and-metrics-db223bd9738a


Acing AI: Lyft’s End-to-End ML Platform

Flyte is the workflow automation platform for complex, mission-Critical Data and ML processes at scale. The blog narrates a general overview of Flyte, integration with data catalog, and extensibility of the platform.

https://medium.com/acing-ai/lyfts-end-to-end-ml-platform-e4498fb1c089


Razorpay: How Razorpay uses Druid for seamless analytics and product insights?

Razorpay writes about its journey of adopting Apache Druid from Apache Kylin & Spark for multi-dimensional analysis. The blog narrates some of the cluster tunings of Druid, how it improves the performance of the data platform, and some of the challenges such as auto-scaling Druid's middle manager, enhance analytics on complex data types.

https://medium.com/@birendra.sahu_77409/how-razorpay-uses-druid-for-seamless-analytics-and-product-insights-364c01b87f1e


Astronomer: Airflow and Ray - A Data Science Story

Ray is a Python-first cluster computing framework that allows Python code, with complex libraries or packages, to be distributed and run on clusters of infinite size. In this blog, Astronomer writes about Airflow integration with Ray using the task flow API and narrates how it uses Ray's in-memory object storage to pass data between the tasks instead of Airflow's traditional XCom approach.

https://www.astronomer.io/blog/airflow-ray-data-science-story


Anna Geller: How a Shared Slack Channel Can Improve Your Data Quality

Integrating the data quality process with the developer workflow and monitoring process is a critical aspect of a data platform's success. The author discusses one such process of integrating data quality alerting and monitoring with Slack and the business process to ensure high data quality standards.

https://towardsdatascience.com/how-a-shared-slack-channel-can-improve-your-data-quality-e62a4c2a0936


Confluent: Kafka Summit Europe 2021 Recap

Confluent writes about a recap of the recent Kafka summit - Europe 2021. Some exciting talks on data mesh foundation, a deep dive on Zookeeper-less Kafka, and the importance of schema registry & structured streaming.

https://www.confluent.io/blog/highlights-from-kafka-summit-europe-2021/

Data Mesh Talk:


Links are provided for informational purposes and do not imply endorsement. All views expressed in this newsletter are my own and do not represent current, former, or future employers' opinions.