1 MIN READKafka · Cost · Observability
3 Kafka anti-patterns that inflate your cloud bill
The most common Kafka setups we see burning money in production — and how to fix them without a rewrite.
Most teams don't have a Kafka performance problem. They have a Kafka cost problem they can't see. Here are three patterns we run into on almost every audit.
1. Over-partitioned topics
Partitions are cheap until they aren't. Thousands of partitions per broker inflate metadata, slow rebalances and waste memory. Right-size to your real throughput and consumer parallelism — not to a number someone picked in 2021.
2. Replication you don't actually need
replication.factor=3 everywhere feels safe. But for derived, replayable or short-lived topics, you may be paying 3x storage and inter-AZ traffic for data you could rebuild. Tier your topics by how much durability they truly require.
3. No cost attribution
If you can't map spend to a team, topic or index, you can't optimize it. Tag and measure first. Visibility alone usually surfaces 20–30% of easy savings.
These are exactly the problems our audits surface in the first week. If your Kafka bill is climbing faster than your traffic, let's talk.
NEXT STEP
Dealing with this in production?
We deploy and run the systems these notes come from — fixed scope, named deliverables, support under SLA.