Suggestion: Track ActiveRecord connection pool stats

[rainerborene]

A suggestion for a new probe on this library by using this piece of code: ActiveRecord::Base.connection_pool.stat

[krystof-k]

I needed to achieve something similar and unfortunately it turned out to be more complicated.

First I was hoping such initializers/appsignal.rb would sort it out:

Appsignal::Probes.register(:active_record_connection_pool, lambda do
  ActiveRecord::Base.connection_pool.stat.each do |key, value|
    hostname = Appsignal.config[:hostname] || Socket.gethostname
    Appsignal.set_gauge("active_record_connection_pool_#{key}", value, {hostname:})
  end
end)

But this doesn't work for Puma, because the probe runs in the main process but the connection pool is managed for each forked process (Puma worker) independently.

Therefore I needed to "create own probe" for Puma workers like this in config/puma.rb (important note: I have preload_app! in the config):

before_fork do
  # Deregister AppSignal probe from the Puma main process
  Appsignal::Probes.unregister(:active_record_connection_pool)
end

on_worker_boot do |worker_index|
  # Send the gauge metric to AppSignal manually
  Thread.new do
    loop do
      sleep Appsignal::Probes::ITERATION_IN_SECONDS
      ActiveRecord::Base.connection_pool.stat.each do |key, value|
        hostname = Appsignal.config[:hostname] || Socket.gethostname
        Appsignal.set_gauge("active_record_connection_pool_#{key}", value, {hostname:, process: worker_index})
      end
    end
  end
end

Creating such probe would probably need deeper changes in how the probes are registered.

Also this is maybe a little related to #937.

[tombruijn]

The data not being in the previous comment was because of preload_app! loading AppSignal in the Puma primary process. There the probes didn't have access to the connection pool data. We added docs for how to configure AppSignal when using this Puma config: https://docs.appsignal.com/ruby/integrations/puma.html#usage-with-preload_app

[krystof-k]

One more update from me. Based on @tombruijn findings, here is my final setup for the ActiveRecord connection pool probe, working both in Puma workers and everywhere else:

intiializers/appsignal.rb rests the same:

Appsignal::Probes.register(:active_record_connection_pool, lambda do
  ActiveRecord::Base.connection_pool.stat.each do |key, value|
    hostname = Appsignal.config[:hostname] || Socket.gethostname
    Appsignal.set_gauge("active_record_connection_pool_#{key}", value, {hostname:})
  end
end)

config/puma.rb is a little cleaner, however it is still required to re-register the probe to include the Puma worker number:

preload_app!

before_fork do
  # Stop AppSignal probes in the Puma main process
  Appsignal::Probes.stop
  # Deregister AppSignal probe from the Puma main process
  Appsignal::Probes.unregister(:active_record_connection_pool)
end

on_worker_boot do |worker_index|
  # Register the AppSignal probe in the Puma worker process
  Appsignal::Probes.register(:active_record_connection_pool, lambda do
    ActiveRecord::Base.connection_pool.stat.each do |key, value|
      hostname = Appsignal.config[:hostname] || Socket.gethostname
      Appsignal.set_gauge("active_record_connection_pool_#{key}", value, {hostname:, process: worker_index})
    end
  end)
  # Start AppSignal probes in the Puma worker process
  Appsignal::Probes.start
end