Pain Point

Schema Evolution

Changing data schemas (adding columns, renaming fields, altering types) in S3-stored datasets without breaking downstream consumers.

16 connections 2 resources 1 post

Summary

What it is

Changing data schemas (adding columns, renaming fields, altering types) in S3-stored datasets without breaking downstream consumers.

Where it fits

Schema evolution is the recurring tension between "business requirements change" and "existing data and queries must keep working." Every table format exists in part to solve this problem.

Misconceptions / Traps
  • Not all schema changes are equal. Adding a column is safe in all table formats; renaming or changing types has format-specific behavior and risks.
  • Schema evolution in the table format does not automatically propagate to downstream tools (dashboards, ML pipelines). Consumer-side schema awareness is still required.
Key Connections
  • Apache Iceberg, Delta Lake, Apache Hudi solves Schema Evolution — table format support
  • Iceberg Table Spec, Delta Lake Protocol, Apache Avro solves Schema Evolution — specification-level solutions
  • Schema Inference solves Schema Evolution — LLM-assisted schema suggestion
  • Write-Audit-Publish catches schema-breaking changes before they reach consumers
  • scoped_to Table Formats, Data Lake

Definition

What it is

The challenge of changing data schemas (adding columns, renaming fields, altering types) in S3-stored datasets without breaking downstream consumers, queries, or pipelines.

Recent developments

Latest signals
  • Iceberg is the strongest schema-evolution story. Iceberg uniquely identifies schema elements with internal IDs, so renames, reorders, and adds work without breaking historical readers — schema changes produce a new metadata.json + commit but no new snapshot. Per Cloudera — Iceberg schema evolution feature.
  • Delta Lake adds type widening + column mapping for in-place evolution. Delta's column-mapping feature allows some schema changes without rewrites; the type widening capability lets you ALTER COLUMN types without rewriting underlying data files. Per Databricks — Delta Lake Type Widening.
  • Hudi supports nullable adds + type promotion out of the box. Hudi handles backwards-compatible schema evolution natively (nullable field add, type promotion); experimental support for backward-incompatible evolution resolves at read time. Per Apache Hudi — Schema Evolution docs.
  • Column drop + rename still requires rewrite across all formats. The universal gotcha in 2026 — even with Iceberg's identifier-based approach, dropping or renaming columns typically requires rewriting data files. Operators commonly mitigate via column mapping + soft-deletes rather than actual physical drops. Per Reintech — Iceberg vs Delta vs Hudi 2026.
  • Three-formats local-laptop comparison published. A 2026 DEV article demonstrates Iceberg + Delta + Hudi in a local laptop setup specifically to compare schema-evolution behavior empirically. Per DEV — Three Formats Walk into a Lakehouse.

Connections 16

Outbound 2
Inbound 14

Resources 2

Featured in