Kafka in ProductionHow Uber, PayPal, Agoda, and 30+ companies actually run Kafka — scale, failure, cost, and the judgment calls.Week 3: The Kafka billCloud Kafka's dominant cost is network, not storage — follower fetching, tiered storage, and the diskless KIPs are one continuous attack on cross-AZ traffic.

Figure 1 — the ladder of attacks on the Kafka bill. By the end of this week you can read this as a prediction: which slice of your own bill each rung removes, and what each one trades away

Finance flags your streaming platform's spend, so you audit it expecting waste: oversized brokers, forgotten topics, bloated retention. You find none. Brokers are right-sized, storage is modest — and the biggest line on the bill is "inter-AZ data transfer," roughly half the total. Nothing is misconfigured. Every one of those bytes is doing exactly what Kafka's design says it should. So where, mechanically, is the money going?This week is the Kafka bill: deriving where the bytes cross zone boundaries, and reading a decade of architecture work — follower fetching, tiered storage, diskless topics — as one continuous attack on that line item.After this week you canThe full week 3 brief is part of LeetData Pro.Case studies

Grab — Zero traffic cost for Kafka consumers.

Pinterest — Pinterest Tiered Storage for Apache Kafka®️ A Broker-Decoupled Approach

Uber — Introduction to Kafka Tiered Storage at Uber

TopicPartition — KIP-1150 in Apache Kafka is a big deal (Diskless Topics)

Fresha — The Good, The Bad, and The AutoMQ

AWS — Best practices for right-sizing your Apache Kafka clusters to optimize performance and cost

Exercises

The AZ Outage Decision

The Cross-AZ Bill

Week 3 check: The Kafka Bill