Rails Caching Patterns Skill

Rails Caching Patterns

Analyze and recommend caching strategies across all layers of a Rails application.

Follow standard Rails conventions for basic cache helper, Rails.cache.fetch, stale?/fresh_when, and expires_in. Focus on the opinionated patterns below.

See patterns.md for detailed code examples.

Quick Reference

| Pattern | Use When | |---------|----------| | Russian doll + touch: true | Nested associations — inner updates bust outer caches | | race_condition_ttl | High-traffic keys — prevents cache stampede on expiry | | fetch_multi | Multiple cache reads — one round-trip instead of N | | Collection caching (cached: true) | Rendering collections — uses read_multi internally | | Solid Cache vs Redis | Match the configured store; default Solid Cache unless Redis is already in the stack | | Composite cache keys | Varying by user, locale, or version |

Choosing a Cache Store

The cache store is an orthogonal project fact, not an architecture choice. If the project already configures one, match it (project-conventions fingerprint). For a new choice, pick by what is already in the stack:

| Store | Use when | Why | |-------|----------|-----| | config.cache_store = :solid_cache_store | No Redis in the stack | No external deps, database-backed, Rails 8 default | | config.cache_store = :redis_cache_store, { url: ENV["REDIS_URL"] } | Redis already running (Sidekiq, ActionCable) | Reuse existing infrastructure | | :memory_store | Development | Toggle with bin/rails dev:cache |

Core Principles

Cache the expensive thing closest to where it's used: HTTP > fragment > low-level
Key-based expiration over manual invalidation: Let cache_key_with_version change when data changes
Measure before caching: Use rack-mini-profiler or server timing — don't cache what isn't slow
Russian doll from outside in: touch: true propagates changes up the association chain
Never cache user-specific content in shared caches without varying the key

Russian Doll Caching

The critical setup: touch: true on associations propagates updated_at changes upward.

class Comment < ApplicationRecord
  belongs_to :post, touch: true
end

class Post < ApplicationRecord
  belongs_to :category, touch: true
  has_many :comments, dependent: :destroy
end

<% cache @category do %>
  <h2><%= @category.name %></h2>
  <% @category.posts.includes(:comments).each do |post| %>
    <% cache post do %>
      <h3><%= post.title %></h3>
      <% post.comments.each do |comment| %>
        <% cache comment do %>
          <p><%= comment.body %></p>
        <% end %>
      <% end %>
    <% end %>
  <% end %>
<% end %>

How it works: Comment updated -> post.touch -> category.touch. Only changed fragments re-render; unchanged siblings hit cache.

Race Condition TTL

Prevents cache stampede when many requests hit an expired key simultaneously. First request recalculates; others get stale data for the TTL window:

Rails.cache.fetch("dashboard/stats", expires_in: 5.minutes, race_condition_ttl: 30.seconds) do
  DashboardStats.calculate  # Expensive
end

Collection Caching with `read_multi`

One cache round-trip for the entire collection:

<%= render partial: "products/product", collection: @products, cached: true %>

`fetch_multi` for Batch Cache Reads

Avoid N cache round-trips:

class Product < ApplicationRecord
  def self.with_cached_stats(products)
    keys = products.map { |p| "product/#{p.id}/stats" }

    cached = Rails.cache.fetch_multi(*keys, expires_in: 1.hour) do |key|
      product_id = key.split("/")[1].to_i
      calculate_stats_for(product_id)
    end

    products.each do |product|
      product.cached_stats = cached["product/#{product.id}/stats"]
    end
  end
end

Composite Cache Keys

# Varies by multiple factors
cache [@product, current_user.locale, "v2"] do ... end

# Conditional caching — skip cache for admins
cache_if !current_user&.admin?, @product do ... end

Write-Through Cache

Update cache proactively when data changes instead of waiting for expiry:

class Product < ApplicationRecord
  after_commit :update_cache, on: [:create, :update]
  after_commit :clear_cache, on: :destroy

  def cached_stats
    Rails.cache.fetch(stats_cache_key, expires_in: 1.day) do
      calculate_stats
    end
  end

  private

  def update_cache
    Rails.cache.write(stats_cache_key, calculate_stats, expires_in: 1.day)
  end

  def clear_cache
    Rails.cache.delete(stats_cache_key)
  end

  def stats_cache_key = "product/#{id}/stats"
end

Anti-Patterns

| Anti-Pattern | Fix | |-------------|-----| | Caching without measuring | Profile first with rack-mini-profiler | | Manual invalidation everywhere | Use key-based expiration (cache_key_with_version) | | No touch: true with Russian doll | Parent cache never busts — add touch: true | | Long expires_in without race_condition_ttl | Cache stampede on expiry — add race_condition_ttl: 30.seconds | | Rails.cache.fetch in a loop | N round-trips — use fetch_multi | | Caching nil results | Thundering herd — return null object or empty result | | User-specific content in shared cache | Data leaks — include user/session in cache key |

Output Format

When recommending caching, provide:

Layer — which caching layer applies (HTTP, fragment, low-level)
Implementation — code with cache keys and expiration
Invalidation strategy — how/when the cache busts
Measurement — how to verify the cache is helping

Agent Skills: Rails Caching Patterns

Install this agent skill to your local

Skill Files