File formats & storage · Formats & Table Formats

What is Delta Lake?

MediumDelta LakeACIDParquettime travellakehouse

Interview question

What is Delta Lake, and what does it add on top of Parquet files?

How to answer

The short answer

Delta Lake is a table format (and transaction log) that turns a directory of Parquet files into an ACID table with versioning. The data is still Parquet; the magic is the _delta_log that records commits.

Problem first. A bare Parquet folder has no safe concurrent writers, no reliable UPDATE/DELETE, and no easy time travel. Readers can see half-written jobs. Delta adds warehouse-like table semantics on cheap .

Output
/table/
  part-000.parquet
  part-001.parquet
  _delta_log/
    000000.json   <- commits: add/remove files, schema, etc.
    000001.json

What Delta gives you

  • ACID transactions (atomic commits of file adds/removes)
  • MERGE / upserts, deletes, updates
  • Time travel (VERSION AS OF / timestamp)
  • Schema enforcement and evolution controls
  • Optimize / Z-Order / vacuum tooling (engine-dependent)

Interview tip: "Delta = Parquet data files + transaction log." Stress that the format, not just Databricks, defines the table; Spark, Trino, and others can read Delta depending on the stack.