Airflow 2 lost open-source security support on April 22, 2026. Read about the migration
Airflow 2 is out of support. Migration details
2026
A fintech running its data pipelines on Astronomer.
Mostly parameter checks. Everything else needed routine changes.
Operator imports updated in every DAG, providers upgraded for Airflow 3, and code that called the old REST API or the metadata database rewritten.
Both versions ran side by side for 2 weeks, with outputs compared in Great Expectations.
Rehearsed first, with Airflow 2 kept ready as a rollback.
2025
A fintech handling customer PII.
Snowflake spend was high for the workload, customer PII had to stay in the client’s own infrastructure, and the product needed low-latency queries a warehouse handles poorly.
Snowflake-only SQL (QUALIFY, FLATTEN, date and string functions) rewritten for Postgres, VARIANT columns and their queries moved to jsonb, and Snowflake-specific macros and packages replaced or adapted.
Row counts and column-level values compared for every model, and the existing dbt tests passed on Postgres.
Platform costs down about 40%, app queries faster, and PII stays in the client’s infrastructure.
2024
A retailer selling in stores and online.
Spark jobs on Amazon EMR over sales, inventory, and e-commerce data, an S3 data lake, and AWS Step Functions scheduling each run.
Seasonality, store clustering, lost sales, and size curves.
Clusters start for each run and shut down after, with task nodes on spot instances, so nothing idles between runs.
Still in production, run by the client’s team.
The call is free. Tell us what you run and where it hurts. If it’s not a fit, we’ll say so.
You’ll talk to John García, a data engineer with over 15 years building production pipelines.
Rather write? Leave a note and we’ll reply by email.