Connect Interview Questions

Use messaging for decoupling, buffering spikes, fan-out, and async workflows. Connect via Kafka trades immediate consistency for scalability—explain when that trade-off is acceptable.

Publisher confirms, durable queues, consumer acknowledgments, and dead-letter queues. Design Connect consumers to be idempotent because at-least-once delivery is common.

When consumers lag, queues grow and memory pressure increases. Apply rate limits, scale consumers, or shed load. Monitor queue depth for Connect pipelines and alert before SLA breach.

Use Avro, Protobuf, or JSON schemas with versioning. Consumers should tolerate unknown fields and evolve schemas compatibly when Connect event shapes change.

Partition keys preserve order per entity; global order is expensive. Design Connect so out-of-order delivery is handled or explicitly ruled out by architecture.

Use compacted topics, replay tools, or shadow consumers in non-prod. Ensure replays do not double-apply side effects unless consumers are idempotent.

Lag, throughput, error rate, rebalance events, and broker disk usage. Dashboards for Connect help catch consumer stalls before messages expire.

Use the official Kafka docs for Connect, language or framework references, and reputable community guides. Bookmark release notes and migration guides when upgrading versions, since Connect behavior can change between releases.

Copying snippets without understanding why Connect works leads to fragile code. Beginners often skip error handling, tests, or edge cases. Slow down, trace execution step by step, and validate assumptions with small experiments.

Start with the problem Connect solves, show a minimal working example, and list the team conventions around it. Point them at official docs and one trusted internal example rather than random snippets.

Connect shows up often in production Kafka work—misunderstanding it leads to bugs, performance issues, or security gaps. Interviewers want clear explanations plus practical judgment.

Describe scaffolding a feature, configuring defaults, or validating input where Connect is required. Call out what goes wrong if the team skips conventions around it.

Read the official overview, run a minimal sandbox, learn key terms and common errors, then expand with a small project. Hands-on practice beats memorizing Connect definitions.

Follow team conventions, split concerns into testable units, handle edge cases, and document assumptions. Review similar modules in the codebase, add observability, and ship incrementally with feature flags if Connect is risky.

Map the Connect attack surface (injection, broken auth, data exposure, DoS). Layer defenses—validation, rate limits, least privilege, encryption, and regular audits.

Measure first, then improve the hottest Connect paths with batching, connection pooling, async I/O, better algorithms, or sharding. Re-check p95/p99 after each change and skip micro-tweaks without clear gains.

Use expand/contract or strangler patterns, dual-write/dual-read where needed, feature flags, and rollback plans. Validate parity with shadow traffic before decommissioning the old Connect path.

State whether Connect needs strong consistency or can tolerate eventual consistency. Discuss partitions, quorum, conflict resolution, and user-visible anomalies during failures.

Pick availability and latency indicators, set achievable objectives, watch burn rate, and decide when reliability work outranks features. Tie those budgets to release decisions for Connect.

Define clear ownership boundaries for Connect, failure domains, caching, and observability. Plan capacity, multi-region needs if relevant, and explicit trade-offs between consistency, latency, and cost.