Retiring Lambda — Week 4 CheckWeek 4 of Stream Processing: unified batch-stream architectures and streaming's boundary with the lakehouse.

The Lambda architecture (parallel batch + speed layers) is being retired at companies like LinkedIn (−94% processing time after unification). What was Lambda's fatal operational flaw?

The Kappa alternative says: one streaming pipeline, and reprocessing = …

Alibaba runs Double-11 on stream-batch unification (4B records/sec peak). The engine-level idea that makes 'unified' real rather than marketing:

Streaming's killer app in the corpus is real-time ML features. The defining correctness problem of feature pipelines:

The streaming-lakehouse boundary (this week's Two-Minute Iceberg scenario) forces a triangle. Its corners:

When does the honest answer become 'you don't need streaming here'?