Buying Kafka Support: The Guide

Broker incidents, configuration and tuning, upgrade guidance and patch advice are in scope on almost every contract. Client code, connectors, Kafka Streams applications and the Kubernetes platform underneath are where contracts differ, and where the real incidents come from. Get the boundary in writing, with examples, before the first ticket.

Apache Kafka is fully supportable as open source; you do not need a Confluent subscription to have a production estate covered. Confluent Platform adds components (Schema Registry, connectors, Control Center, tiered storage) and a support tier that covers them. The question is which components you actually run, because that decides whether vendor support covers your incident or your broker's.

Vendor support follows the product: it is authoritative on the distribution and stops at its edge. Independent support follows the estate: self-managed Apache Kafka, Amazon MSK and Confluent Platform under one contract, with the client and connector layer in scope where the contract says so. Most estates that run more than one of those end up with independent cover, sometimes alongside a vendor subscription for the components only the vendor can patch.

Priced on cluster and broker count plus the response tier, from business hours through 24/7 with a 15-minute emergency response. Self-managed estates with a support contract usually cost less per year than a fully managed service and more than nothing, and the comparison that matters is against the cost of the last outage and the engineer time spent on the next upgrade.

Four things decide whether an SLA is worth anything: what starts the clock (ticket logged, or engineer engaged), who the first responder is (a senior engineer or a triage tier), what severity definitions say about a broker that is up but lagging, and whether the response commitment covers the platform under the broker. A 15-minute response from a triage queue is not a 15-minute response.

Which clusters and which components, whether client code and connectors are in scope, whether MSK or Confluent Cloud incidents are covered when the platform is the cause, how upgrades are handled, and what a health check at onboarding produces. Every dispute in a support relationship sits on one of those lines; draw them first.