When the Java Virtual Machine (JVM) starts, it starts by interpreting Java bytecode. As it starts to detect code that runs often, it just-in-time compiles that code into native machine code, improving performance.

This is a known challenge for most JVMs, and the same applies to JRuby applications, which also run on the JVM.

A widely-documented solution to this problem is to perform a warm-up step when starting a service:

• https://landing.google.com/sre/book/chapters/load-balancing-datacenter.html#unpredictable-performance-factors-JMs7i7trCj

• http://www.brendangregg.com/blog/2016-09-28/java-warmup.html

• https://devcenter.heroku.com/articles/warming-up-a-java-process

WarmBlanket warms services by performing repeated web requests for a configurable number of seconds. After that time, it closes shop and you’ll never hear about it until the next service restart or deploy.

WarmBlanket spawns a configurable number of background threads that run inside the service process, and then uses an http client to perform local requests to the web server, simulating load.

As it simulates requests, the JVM is warmed up and thus when real requests come in, no performance degradation is observed.

We strongly recommend that any services using WarmBlanket, if deployed on Heroku, use Preboot. Preboot allows a service instance to be warmed up for 3 minutes before Heroku starts sending live traffic its way, which is preferable to doing it live.

On kubernetes, you can make use of readiness probes to delay service startup while warm-blanket is working.

TruffleRuby can be run either on a standalone mode, or with a JVM. In either configuration, like JRuby, it needs to warm up until it gets to peak performance, so WarmBlanket is great there too!

To install using Bundler, add the following to your Gemfile:

WarmBlanket uses semantic versioning.

This gem can be configured via the following environment variables:

• PORT: Local webserver port (automatically set on Heroku)

• WARMBLANKET_ENABLED: Enable warm blanket (defaults to false; true or 1 enables)

• WARMBLANKET_WARMUP_THREADS: Number of warm up threads to use (defaults to 2)

• WARMBLANKET_WARMUP_TIME_SECONDS: Time, in seconds, during which to warm up the service (defaults to 150)

Configuring endpoints to be called

require 'warm-blanket'

WarmBlanket.configure do |config|
  common_headers = {
    'X-Api-Key': ENV.fetch('API_KEYS').split(',').first,
  }

  config.endpoints = [
    {get: '/foo', headers: common_headers},
    {get: '/', headers: common_headers},
  ]
end

# Notice that you need to both:
# * set the Content-Type manually (if needed)
# * JSON-encode the body  (if needed)

WarmBlanket.configure do |config|
  common_headers = {
    'X-Api-Key': ENV.fetch('API_KEY').split(',').first,
    'Content-Type': 'application/json',
  }

  post_body = MultiJson.dump(
    account_id: 'dummy_account',
    user_id: 'dummy_user_id',
  )

  config.endpoints = [
    {post: '/some_endoint', headers: common_headers, body: post_body},
  ]
end

Add the following to the end of your config.ru file: