Kafka Schema Registry Compatibility Governance Prompt
Design and enforce a schema evolution policy in Confluent/Apicurio Schema Registry so producers and consumers stay compatible through changes without breaking downstream consumers.
- Target user
- Platform and backend engineers
- Difficulty
- Advanced
- Tools
- Claude, ChatGPT, Cursor
The prompt
You are a senior Kafka platform engineer defining schema-evolution governance for a Schema Registry, producing a compatibility policy and a safe change-review checklist to apply before schemas are registered. I will provide: - Registry setup: Avro/Protobuf/JSON Schema, current global and per-subject compatibility level (BACKWARD, FORWARD, FULL, or *_TRANSITIVE, NONE) - Subject naming strategy: TopicNameStrategy vs. RecordNameStrategy vs. TopicRecordNameStrategy - A proposed schema change (added/removed/renamed fields, type changes, default changes) and the affected topic - Deployment order reality: whether producers or consumers deploy first, and whether old consumers linger Your job: 1. **Match compatibility to deploy order** — recommend BACKWARD when consumers upgrade first, FORWARD when producers upgrade first, and FULL when order is uncontrolled, explaining the trade-off for this pipeline. 2. **Evaluate the proposed change** — determine whether adding a field (needs default), removing a field, renaming, or narrowing a type passes the chosen level, and flag exactly which rule it violates. 3. **Decide on TRANSITIVE** — advise whether transitive checking against all prior versions is warranted given how long old messages and old consumers persist. 4. **Fix unsafe changes** — rewrite the schema to be compatible (add defaults, keep old fields as deprecated, use unions/aliases) rather than forcing a breaking bump. 5. **Handle unavoidable breaks** — when a break is genuinely required, prescribe the versioned-topic or dual-write migration path instead of loosening compatibility to NONE. 6. **Enforce in CI** — specify a pre-merge compatibility check (Maven/Gradle plugin or registry test endpoint) so incompatible schemas fail the build, not production. Output: (a) recommended per-subject compatibility level with rationale, (b) verdict on the proposed change with the specific rule it hits, (c) a compatible rewrite or a migration plan, (d) a CI enforcement snippet. Advisory only; validate every schema against the registry's test-compatibility endpoint in a non-prod registry before registering in production.
Run this prompt with AI
Test it, get an AI-improved version, or compare models — live in the Prompt Workspace. No copy-paste.
Related prompts
-
Kafka Multi-Tenancy and Topic Naming Governance Prompt
Define a topic naming convention, ACL model, and quota boundaries for a shared multi-tenant Kafka cluster so tenants are isolated, discoverable, and enforceable by prefix.
-
Kafka Rolling Version Upgrade Plan Prompt
Plan a zero-downtime rolling upgrade of a Kafka cluster across versions, sequencing inter.broker.protocol.version and log.message.format.version bumps so clients never break mid-upgrade.
-
Kafka Client Quota and Throttling Design Prompt
Design produce, fetch, and request-percentage quotas per user/client-id so one noisy tenant cannot saturate broker network or CPU and starve others on a shared cluster.
-
Kafka Connect Dead Letter Queue and Error Handling Design Prompt
Design connector-level error tolerance, dead-letter-queue routing, and retry/backoff for Kafka Connect so a single poison record or transient sink failure never silently drops data or stalls the connector.
More Kafka prompts & error guides
Browse every Kafka prompt and troubleshooting guide in one place.
Reading prompts? Get all 500 in one free PDF
500 battle-tested, copy-paste AI prompts engineered by a senior systems engineer — every one with fill-in placeholders and safety/back-out notes. Drop your email and it's yours.
- 500 prompts: Linux · Kubernetes · Terraform · OpenStack · GitLab · Docker · Monitoring · Incident Response
- Instant PDF download — yours free, forever
- Plus one practical AI-workflow email a week (no spam)
Single opt-in · unsubscribe anytime · no spam.