SkillVaultskills Browse all 1,000+ skills

Performance · Version 1.2.0 · Reviewed 2026-08-02

Apache Kafka Performance Tuning Specialist

Locate and remove the dominant bottleneck in Apache Kafka latency attribution and Apache Kafka throughput optimization with evidence, explicit trade-offs, and a verification plan.

4 method steps 6 documented failure modes 5 diagnostic checks 7 quality gates

Finds the dominant measured bottleneck and designs representative benchmarks for Apache Kafka using topic configuration, partitioning, producer settings, and consumer groups and per-partition lag, rebalance history, under-replicated partitions, and request latency, with explicit attention to key skew or rebalance churn stalling one partition while aggregate metrics look healthy.

₹199 one-time

Get this skill archive

Install in your AI coding tool

SkillVault packages this skill in the open Agent Skills format for five leading coding tools.

What this skill helps you do

  • Apache Kafka latency attribution
  • Apache Kafka throughput optimization
  • Apache Kafka performance regression guard

How Apache Kafka Performance Tuning Specialist works

You provide

Topic config, consumer settings, and lag history

It inspects

Per-partition lag and processing time for Apache Kafka latency attribution

It decides

A Apache Kafka throughput optimization decision with ordering guarantees stated

You verify

Lag flattens and dead-letter volume stays bounded

What it checks first

Apache Kafka Performance Tuning Specialist finds the dominant measured bottleneck and designs representative benchmarks for Apache Kafka using topic configuration, partitioning, producer settings, and consumer groups and per-partition lag, rebalance history, under-replicated partitions, and request latency, with explicit attention to key skew or rebalance churn stalling one partition while aggregate metrics look healthy. Use it when the work involves Apache Kafka latency attribution, Apache Kafka throughput optimization, Apache Kafka performance regression guard.

  1. Consumer lag trend rather than absolute value: flat lag at any level is healthy, rising lag is not.
  2. Partition count versus consumer count, since consumers beyond the partition count are idle by definition.
  3. Whether the partition key produces even distribution, or a few keys dominate one partition.
  4. Rebalance frequency, which converts into repeated processing pauses.
  5. Whether offsets commit before or after processing, which decides between at-most-once and at-least-once.

Failure modes it recognizes

  • A poison message blocking a partition indefinitely because the consumer retries in place without dead-lettering.
  • Processing time exceeding the poll interval, causing the broker to evict the consumer and trigger a rebalance loop.
  • Committing offsets before processing, silently dropping messages on crash.
  • Producer key set to null or a timestamp, destroying ordering guarantees the consumer assumed.
  • Consumer group rebalance storms from short session timeouts on a slow consumer.
  • Unbounded retention plus a compaction misconfiguration filling disk and stopping the broker.

Answers it will reject

  • Adding consumers to reduce lag when partitions are already saturated — throughput is bounded by partitions.
  • Increasing partitions to fix lag without checking whether processing is CPU-bound downstream.
  • Treating the queue as a database by retaining everything and querying it by scan.
  • Requeueing a failed message to the tail forever, converting a bug into an infinite loop.

Decision rules it applies

  • Order is guaranteed only within a partition, so any ordering requirement must map to a partition key.
  • Choose at-least-once with idempotent consumers over attempting exactly-once across systems.
  • Dead-letter after a bounded retry count with the failure reason attached; never retry indefinitely in place.
  • Size partitions for peak throughput plus headroom, because increasing partitions later breaks key-to-partition mapping.

Evidence it asks for

  • Track consumer lag per partition, not aggregated, so a single stuck partition is visible.
  • Measure processing time per message against `max.poll.interval.ms`.
  • Alert on rebalance rate and on dead-letter volume as separate signals.

The method inside

  1. Extract decisions, facts, and unresolved questions needed for Apache Kafka latency attribution.
  2. Organize Apache Kafka throughput optimization around the reader's next decision or action rather than the source order.
  3. Draft Apache Kafka performance regression guard with source traceability and no invented behavior.
  4. Run a completeness, consistency, audience, and actionability review before returning the artifact.

Deliverables

  • Apache Kafka latency attribution assessment
  • Apache Kafka throughput optimization decision and action plan
  • Apache Kafka performance regression guard verification checklist

Evidence requirements

  • Profiles, traces, timings, resource metrics, and workload shape
  • Baseline and target percentile
  • Environment, concurrency, and payload details

Quality gates

  • Every material claim traces to supplied evidence or is labeled as a hypothesis.
  • The response follows the declared deliverable contract.
  • No execution, access, measurement, or verification is invented.
  • Secrets and personal data are redacted rather than repeated.
  • The user receives a concrete independent verification step.
  • The relevant failure modes in this domain were considered rather than only the reported symptom.
  • No listed anti-pattern was recommended as a solution.

Example task

Input

Apply the performance tuning specialist to our Apache Kafka system before the next production change. We can provide topic configuration, partitioning, producer settings, and consumer groups; the main concern is key skew or rebalance churn stalling one partition while aggregate metrics look healthy.

Expected output

Define the failing percentile and workload, then attribute time with per-partition lag, rebalance history, under-replicated partitions, and request latency. The likely mechanism to disprove first is key skew or rebalance churn stalling one partition while aggregate metrics look healthy. Change one constraint at a time and compare resource use, tail latency, and correctness against a pinned baseline.

Boundaries and compatibility

Ideal for

  • Apache Kafka latency attribution: produce a decision or artifact grounded in supplied evidence.
  • Apache Kafka throughput optimization: produce a decision or artifact grounded in supplied evidence.
  • Apache Kafka performance regression guard: produce a decision or artifact grounded in supplied evidence.

Out of scope

  • Optimizing without a baseline
  • Using averages where tail latency determines experience

Agent compatibility

  • GitHub Copilot Agent Skills
  • Cursor Agent Skills
  • Claude Code Skills
  • OpenAI Codex Skills
  • JetBrains Junie Skills

Tool policy: Advisory by default. No tools are assumed. If the host provides tools, use read-only evidence gathering unless the user explicitly approves a scoped write or execution action.