yabeda-rack-queue
March 9, 2026 ยท View on GitHub
Rack middleware that measures HTTP request queue time. It reports the result to Yabeda as a histogram.
What is queue time?
A request may wait before your app handles it. A proxy or load balancer (like Nginx or Heroku) causes this wait. This is called queue time.
High queue time means your app is too busy. It cannot take new requests. That is a sign you need more capacity.
How it works
Load balancers can add a header to each request. The header records when the request arrived. Common headers are X-Request-Start and X-Queue-Start.
This middleware reads that header. It subtracts the header's timestamp from the current time. Then it reports that value as rack_queue.duration.
Note
If neither header is present, no measurement is taken. The request passes through unchanged.
Installation
Add to your Gemfile:
gem "yabeda-rack-queue"
Then run:
bundle install
Usage
Add the middleware to your Rack stack. You also need a Yabeda adapter. For example, use yabeda-prometheus.
require "yabeda/rack/queue"
require "yabeda/prometheus"
Yabeda.configure!
use Yabeda::Rack::Queue::Middleware
run MyRackApp
For Rails, add it in config/application.rb:
config.middleware.use Yabeda::Rack::Queue::Middleware
Metric
| Name | Group | Type | Unit |
|---|---|---|---|
duration | rack_queue | histogram | seconds |
Access it in code:
Yabeda.rack_queue.duration
With adapters that prepend the group name, such as Prometheus, this becomes rack_queue_duration_seconds.
Histogram buckets: 1 ms, 5 ms, 10 ms, 25 ms, 50 ms, 100 ms, 250 ms, 500 ms, 1 s, 2.5 s, 5 s, 10 s, 30 s, 60 s.
Header formats
The middleware checks X-Request-Start first. If that header is absent, it tries X-Queue-Start.
Supported timestamp formats:
| Format | Example |
|---|---|
| Seconds (float) | 1609459200.123 |
| Milliseconds | 1609459200123 |
| Microseconds | 1609459200123456 |
t= prefix | t=1609459200.123 |
The middleware auto-detects the unit. It checks if the number fits a valid recent time.
Puma adjustment
Puma sets puma.request_body_wait (in milliseconds) in the Rack env. This records how long Puma spent reading the request body.
The middleware subtracts this value from queue time. Without this step, large bodies make queue time appear too long.
Configuration
The middleware accepts these keyword arguments:
| Argument | Default | Purpose |
|---|---|---|
reporter: | YabedaReporter.new | Writes the value to Yabeda. |
logger: | stderr | Gets warning messages. |
clock: | Process.clock_gettime(CLOCK_REALTIME) | Returns current time in seconds. |
Example with a custom logger:
use Yabeda::Rack::Queue::Middleware, logger: Rails.logger
Requirements
- Ruby >= 3.1.
- yabeda >= 0.14, < 1.0.
- A Yabeda adapter. For example: yabeda-prometheus.
Development
Run tests:
bundle exec rake test
Run the linter:
bundle exec standardrb
Contributing
Bug reports and pull requests are welcome at https://github.com/speedshop/yabeda-rack-queue.
License
MIT. See LICENSE.txt.