Task does not fail if unicorn fails starting because of code error

[oliviermilla]

Hello,
I had a missing require in my app that made unicorn throw an error and stop during cap production deploy. Shouldn't this stop the deploy process?

[sds]

We've run into this issue as well. If the new Unicorn master fails to start for any reason, the unicorn:restart task will not raise an exception, and so the deploy will appear to be successful but in actuality the master hasn't changed.

A workaround we came up with for this was to have our deploy:restart task check every second (up to some timeout) to see if the master pid has changed. Once it has, we know the new master started successfully since we have preload_app set to true.

Some extracted code to illustrate:

namespace :deploy do
  def wait_until(timeout:, check_interval: 1, &block)
    require 'timeout'

    Timeout.timeout(timeout) do
      loop do
        return if block.call
        sleep check_interval
      end
    end
  end

  task :restart do
    on roles(:app), in: :sequence do |host|
      old_pid = capture(:cat, fetch(:unicorn_pid))

      invoke 'unicorn:restart'

      # Wait until the Unicorn master PID changes before moving on
      begin
        wait_until(timeout: 30) do
          # There is a brief period of time where the unicorn pid file is
          # deleted; ignore it and assume it is still there (hence `rescue old_pid`)
          current_pid =
            capture(:cat, fetch(:unicorn_pid)) rescue old_pid
          current_pid != old_pid
        end
      rescue Timeout::Error
        error "Unicorn master did not change"
        exit 1
      end
    end
  end
end

I would be happy to submit a pull request to enable this functionality directly in the unicorn:restart task via some setting (e.g. set :unicorn_check_master_pid_on_restart), but only if this aligns with the goals of the project. @mlineen, do you have any thoughts?

[surfacedamage]

I've also been bitten by this issue. Thank you for the suggestion, @sds.