Skip to content

Ent Historical Metrics

Mike Perham edited this page Feb 13, 2023 · 25 revisions

Sidekiq Enterprise has the ability to send queue processing metrics to Statsd for dashboards and historical reporting.

Configuration

Configure a Statsd client

See the Pro Metrics page for how to tell Sidekiq about your Statsd server.

Enable History

In your initializer, add this:

Sidekiq.configure_server do |config|
  # history is captured every 30 seconds by default
  config.retain_history(30)
end

Metrics

Sidekiq Enterprise sends the following metrics:

  • sidekiq.processed - Number of job executions completed (success or failure)
  • sidekiq.failures - Number of job executions which raised an error
  • sidekiq.enqueued - Total Size of all known queues
  • sidekiq.retries - Total Retries Size
  • sidekiq.dead - Total Dead Size
  • sidekiq.scheduled - Total Scheduled Size
  • sidekiq.busy - Total Busy Size
  • sidekiq.queue.size (with tag queue:#{x}) - Current Size of queue x
  • sidekiq.queue.latency (with tag queue:default) - Latency of the Default Queue

The statsd namespace will be prepended to each metric, e.g. "myapp.sidekiq.busy".

Custom

Notice above that latency is not gathered by default for every queue because it is a relatively expensive operation. You can add custom history metrics by passing a block to retain_history which collects more metrics, including any further important queue latencies. Here we are adding the latency for the bulk and critical queues:

Sidekiq.configure_server do |config|
  config.retain_history(30) do |s|
    s.batch do |b|
      %w(bulk critical).each do |qname|
        q = Sidekiq::Queue.new(qname)
        b.gauge("sidekiq.queue.latency", q.latency, tags: ["queue:#{qname}"])
      end
    end
  end
end

The block will be passed a Statsd instance which quacks like the normal statsd client, e.g. dogstatsd-ruby.

Clone this wiki locally