Skip to content

benchmarks.md

Latest commit

 

History

History
90 lines (69 loc) · 3.06 KB

benchmarks.md

File metadata and controls

90 lines (69 loc) · 3.06 KB
Raw

Benchmarks

This page defines the benchmark plan for measuring Laravel RabbitMQ worker behavior with native ext-amqp.

The goal is not to publish synthetic numbers without context. The goal is to make performance claims reproducible.

What to measure

Measure each scenario with the same PHP, Laravel, RabbitMQ, CPU, memory, and container limits.

Scenario Metric
Publish normal jobs jobs per second, CPU, memory
Consume normal jobs in poll mode jobs per second, CPU, memory
Consume normal jobs in consume mode jobs per second, CPU, memory
Publish with confirms jobs per second, confirm latency
Delayed jobs publish latency, delivery accuracy
Quorum queues throughput and recovery behavior
Worker runtime for 1 hour memory growth and reconnect behavior
Worker runtime for 6 hours memory growth and stability

Suggested comparison set

  • iamfarhad/laravel-rabbitmq with native ext-amqp.
  • A pure PHP AMQP Laravel queue driver.
  • Laravel Redis queue as a familiar baseline.
  • Laravel database queue as a low-throughput baseline.

Environment template

Record this information for every benchmark run:

PHP version:
Laravel version:
RabbitMQ version:
ext-amqp version:
OS or Docker image:
CPU:
Memory:
Queue type:
Worker mode:
Prefetch count:
Publisher confirms:
Message payload size:
Job handler duration:

Benchmark commands

The package ships no benchmark command. The commands below are a sketch of what to build in your own application, not commands you can run:

# Illustrative only — not provided by this package.
php artisan rabbitmq:bench:publish --queue=bench --jobs=10000 --payload-size=1024
php artisan rabbitmq:bench:consume --queue=bench --jobs=10000 --consume-mode=poll
php artisan rabbitmq:bench:consume --queue=bench --jobs=10000 --consume-mode=consume

In the meantime, use a demo application that dispatches a fixed number of jobs and measures end-to-end time. Compare poll and consume modes at the same RABBITMQ_PREFETCH_COUNT, since prefetch only applies to consume mode and is the single biggest lever on consume-mode throughput.

Reporting format

Use a table like this once measurements are available:

Driver Mode Queue Jobs Payload Jobs/sec Peak memory CPU notes
iamfarhad/laravel-rabbitmq consume classic 10,000 1 KB TBD TBD TBD
iamfarhad/laravel-rabbitmq poll classic 10,000 1 KB TBD TBD TBD

Interpretation rules

  • Do not compare results from different machines.
  • Run every benchmark at least three times.
  • Report median values.
  • Include configuration with the result.
  • Separate publish performance from consume performance.
  • Separate classic queues from quorum queues.
  • Report publisher confirms separately because they intentionally add latency.

Roadmap

  • Add benchmark Artisan commands.
  • Add a Docker Compose benchmark environment.
  • Publish first benchmark results for RabbitMQ 3.13 and 4.x.
  • Add a GitHub Actions manual workflow for benchmark smoke tests.