skills.rest
sandeepmvl
sandeepmvlCOMMUNITY
Explore all skills by this author →
GitHub

puma-tuning-and-concurrency

Configure Puma worker and thread sizing for Rails 8 production deployments.

21|2|Updated May 24, 2026
Search Tags:#performance optimization#rails#yjit#jemalloc#concurrency tuning#puma#puma worker killer
One-click install
npx skills add https://github.com/sandeepmvl/rails-skills --skill puma-tuning-and-concurrency
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: puma-tuning-and-concurrency
Source: https://github.com/sandeepmvl/rails-skills/tree/main/skills/27-puma-tuning-and-concurrency
Command: npx skills add https://github.com/sandeepmvl/rails-skills --skill puma-tuning-and-concurrency

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Misconfigured Puma is the most common Rails performance issue after N+1 queries. Default Puma settings are overly conservative, leading to wasted memory, poor concurrency, and missed throughput gains for production Rails applications.

Core Features & Use Cases

  • Workload-aware sizing formulas: Calculates optimal worker count (CPU cores × 1.5 for I/O-bound, ×1 for CPU-bound) and thread count per worker (3-5 default, up to 10 for I/O-heavy workloads) based on your server specs and app type.
  • Memory optimization integration: Guides setup of jemalloc, MALLOC_ARENA_MAX=2, and YJIT for Ruby 3.3+ to cut memory usage by 20-40% and boost throughput by 10-25%.
  • Production-safe configuration: Provides copy-on-write compatible Puma config with preload_app!, fork-safe on_worker_boot hooks for DB/cache reconnection, and PumaWorkerKiller setup to bound memory bloat from leaks.
  • Use Case: A team running a 4-CPU I/O-bound Rails 8 monolith uses this skill to configure 6 Puma workers with 5 threads each, enabling 30 concurrent requests with a ~2.4GB total memory budget, avoiding overprovisioning and underutilization.

Quick Start

Use the puma-tuning-and-concurrency skill to calculate the optimal Puma worker count, thread count, and memory settings for your production Rails 8 deployment on a 4-CPU I/O-bound server.

Frequently Asked Questions about puma-tuning-and-concurrency

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I calculate optimal Puma worker and thread counts for Rails production?▼

Puma worker count is calculated by multiplying CPU cores by 1.5 for I/O-bound workloads or 1 for CPU-bound workloads. Configure 3-5 threads per worker by default, scaling up to 10 threads for I/O-heavy Rails applications to maximize concurrency.

Why does my Rails application use so much memory with default Puma settings?▼

Default Puma configurations lack copy-on-write memory optimization and leak mitigation. Integrating jemalloc, setting MALLOC_ARENA_MAX=2, and configuring PumaWorkerKiller bounds memory bloat, cutting Rails production memory usage by 20-40%.

Do I need YJIT and jemalloc to improve Rails Puma throughput?▼

YJIT and jemalloc are recommended for Ruby 3.3+ Rails deployments to boost Puma throughput by 10-25%. YJIT accelerates Ruby execution while jemalloc optimizes memory allocation, significantly enhancing overall production performance.

What's the best way to configure fork-safe database connections in Puma workers?▼

Use preload_app! for copy-on-write compatibility and implement fork-safe on_worker_boot hooks in your Puma config. This ensures database and cache connections are properly reconnected after Puma workers fork, preventing connection leaks in Rails production.

Can I use PumaWorkerKiller to bound memory leaks in Rails 8 applications?▼

PumaWorkerKiller is configured to bound memory bloat from leaks in Rails 8 applications. It automatically restarts Puma workers that exceed defined memory limits, maintaining stable production memory usage without manual intervention.

Related Skills

Explore More Software Engineering Skills→
rodmena-limitedrodmena-limited
3

moai-lang-ruby

Automate Ruby 3.3+ code generation, testing, and quality validation for Rails 7.2 and RSpec.

Official
Advanced
arbazkhan971arbazkhan971
26

perf

Profile CPU, memory, and concurrency bottlenecks in Node.js, Python, Go, Rust, and Java services.

Community
Advanced
hung-phanhung-phan

connection-pooling

Configure and monitor SQL database connection pools to reduce connection overhead.

Community
Advanced
6luciano36luciano3

performance-optimization

Profile and optimize Python and Rust code for latency and memory usage.

Community
Advanced