Home › Learn › Real-time analytics › Lambda vs Kappa
Real-time analytics · page 2 of 11
Lambda vs Kappa, drawn simply
Lambda runs a batch path and a speed path and merges them; Kappa uses one replayable log. New builds pick Kappa with a lakehouse for history.
In plain words
Two paths or one
Lambda runs two paths side by side. A batch path recomputes correct history slowly, a speed path gives fresh but rough numbers, and a serving layer merges the two. Kappa keeps one path. Every change is an event in a log you can replay; to fix the logic, you replay the log through the new code.
Side by side
How they compare
| Lambda | Kappa | |
|---|---|---|
| Idea | A batch path for correct history, a speed path for fresh numbers, merged at the end | One path. Everything is an event in a replayable log. |
| Code | The same logic written twice, and the two copies drift apart | Logic written once |
| Fixing history | The batch job recomputes it | Replay from the log, or from lakehouse tables when the log keeps only a few days |
| When it fits | An older setup where the batch path already exists and must stay | Almost every new build |
| Verdict | Drop, unless you inherit it | Pursue |
Today
Kappa with a lakehouse underneath
Modern Kappa also writes the same events to a lakehouse, as Iceberg or Delta tables. That gives you cheap history and a second place to replay from when the log only keeps a few days. The lakehouse is a second destination, not a second logic path, so the design is still Kappa.
A data platform built on the medallion layers can do the same thing: run the same layers at two speeds, with hot tables flowing continuously. See the streaming lane of the data platform guide.
FAQ
Questions people ask
What is the difference between Lambda and Kappa architecture?
Lambda runs a batch path and a speed path and merges them, so the logic is written twice. Kappa sends everything through one replayable log and one processing path, and fixes history by replaying the log.
Is Lambda architecture dead?
For new builds, mostly yes. It survives where a batch path already exists and must stay. New designs use Kappa, usually with a lakehouse for cheap history.
How does Kappa reprocess history if Kafka only keeps 7 days?
It replays from the lakehouse copy (Iceberg or Delta), which keeps everything cheaply, or from Kafka's tiered storage if you keep a longer log.
Next
Where to go next
Next page: The components, left to right
↑ All of real-time analytics · Data platform architecture · Key concepts