How Delta Lake Achieves ACID Using Only JSON Files

The _delta_log commit protocol in 10 minutes

Every Databricks data engineer knows Delta Lake gives them ACID transactions. Most can’t explain how.

There’s no database server running. No write-ahead log. No lock manager. Just Parquet files on cloud storage.

So how does a 10TB table with 50 concurrent writers guarantee atomicity, consistency, isolation, and durability?

The answer is a folder called _delta_log/ and a commit protocol built entirely on numbered JSON files.

Already familiar with the basics? Jump to “What Happens During Concurrent Writes.”