The simplification
Lambda architecture forces you to maintain batch and streaming logic side by side. Kappa architecture removes the batch layer entirely and runs a single streaming pipeline. There is one code path, not two.
How it works
- All data flows into a durable replayable log, such as Kafka with long retention.
- A single stream processor reads the log and produces serving views.
- To recompute, you do not run a separate batch job. You replay the log from the start through a new version of the stream job.
Why this is appealing
- One implementation: the same code computes both live and historical results, so no logic drift.
- Reprocessing is normal: a logic change means spinning up a fresh consumer that reads the log from offset zero into a new view, then cutting over.
What it requires
- The log must retain enough history to replay, which can be large and costly.
- The stream processor must be fast enough that a full replay finishes in reasonable time.
Key idea
Kappa architecture treats batch as just a replay of the stream, so a single streaming codebase over a durable log handles both real time and historical recomputation.