Can you build observability ingestion on S3 alone — no Kafka, no disks, no coordination layer?

We replaced a Kafka + Flink + OTel pipeline (~$700–800/mo at 10 MB/s) with one engine where the data, the WAL, and the Apache Iceberg catalog all live in S3 — no stateful layer to operate. Here's how it works.

Raccontata dadev.to