Reports on thread backlog in the web server
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
bin
doc/images
lib
spec
.gitignore
.rspec
.travis.yml
CHANGELOG.md
Gemfile
LICENSE
README.md
Rakefile
backlog_reporter.gemspec
jenkins.sh

README.md

Backlog Reporter

This gem monitors the size of the web server internal request backlog. (The backlog is requests whose socket has been accepted, but which are not yet being processed by the web server.) Once the request backlog exceeds a certain threshold, a flag file is touched. This flag file can be monitored by a fronting web server (e.g. Nginx) to proactively return 503 Service Unavailable, so that the backlog does not grow further. Once the request backlog falls back below the threshold, the flag file is deleted.

The rapid 503 response prevents the backlog from growing without bound, and it prevents the server from trying to process too many requests in parallel. When the server is trying to process too many requests at once, the average latency starts to climb and all clients start to see longer and longer response times.

Requirements

Currently, backlog_reporter works only with the Puma web server.

Example

Here's an example of a web service whose behavior was improved with the Backlog Reporter.

Without Backlog Reporter

At the 10 request per second rate, the response time starts to degrade until it climbs above 20 seconds per request and the remainder of the requests time out.

Before backlog reporter

With Backlog Reporter

Once the server starts to become overloaded, it degrades in a predictable way. The average response time increases to about 5 seconds; however, it does not increase further from this level. With client retry, all requests will eventually be served successfully.

After backlog reporter

(Note: The performance of this application was also improved at the same time; that's why even the initial responses are faster).

Installation

Add this line to your application's Gemfile:

gem 'backlog_reporter'

And then execute:

$ bundle

Or install it yourself as:

$ gem install backlog_reporter

Usage

In rails config:

  config.backlog_reporter.flag_path = '/run/app/congested.flag'
  config.backlog_reporter.max_backlog = 16
  config.backlog_reporter.check_interval = 0.01 # seconds

In nginx config:

  location / {
    proxy_pass http://localhost:300;

    if (-f /run/app/congested.flag) {
      return 503;
    }
  }

Development

After checking out the repo, run bin/setup to install dependencies. Then, run rake spec to run the tests. You can also run bin/console for an interactive prompt that will allow you to experiment.

To install this gem onto your local machine, run bundle exec rake install. To release a new version, update the version number in version.rb, and then run bundle exec rake release, which will create a git tag for the version, push git commits and tags, and push the .gem file to rubygems.org.

Contributing

Bug reports and pull requests are welcome on GitHub at https://github.com/conjurinc/backlog-reporter.