The Cross-AZ BillKafka · medium · ~14 minFinance flagged the streaming platform: cloud spend doubled in six months while event volume grew only 40%. The line item is inter-AZ data transfer — it's now roughly half the platform's total cost.The cluster is healthy. Nothing is broken. The bill is a *design property* — your job is to explain why, quantify it, and pick the fix that fits an 800ms p99 SLO.Evidence

Metrics

Config

Topology

Checkpoints

Nothing is misconfigured in the classic sense — so why is cross-AZ transfer eating half the bill?

With leaders evenly spread across 3 AZs and consumers placed randomly, what percentage of consumer fetches cross an AZ boundary? (whole number, %)

Which mitigation fits the constraints (800ms p99 SLO, durability non-negotiable, 44 growing consumer groups)?

You've picked follower fetching. Write the rollout plan and name what changes operationally — including who should NOT be moved onto it.