deploy:check failing does not trigger deploy:failed or deploy:finished

[vpillinger]

Steps to reproduce

Add this to a default capistrano project. Do not create .env on the server.

set :linked_files, fetch(:linked_files, []).push('.env')

namespace :deploy do
  desc 'Ending Hook'
  task :ending_hook do
      on roles(:all) do |host|
        puts'ending'
      end
  end
end
before 'deploy:finished', 'deploy:ending_hook '
after 'deploy:failed', 'deploy:ending_hook '

Edit: Fixed rending_hook typo
Then run deploy command.

Expected behavior

The deploy will fail at deploy:check for .env file, then the ending hook will be fired for either deploy:failed or deploy:finished.

Actual behavior

The build stops but no ending hooks are fired.

System configuration

Environment

Ruby     ruby 2.5.1p57 (2018-03-29 revision 63029) [x86_64-linux-gnu]
Rubygems 2.7.6
Bundler  1.16.2
Command  /usr/local/bin/cap staging doctor

Gems

capistrano            3.6.0  (update available)    capistrano            3.6.0  (update available)
airbrussh             1.3.0
rake                  12.3.1
sshkit                1.16.1
net-ssh               5.0.2
capistrano-composer   0.0.6
capistrano-env-config 0.3.0
capistrano-harrow     0.5.3
capistrano-ssh-doctor 1.0.0
capistrano-wpcli      0.1.4

[will-in-wi]

Is it intentional that the before and after definitions involve rending_hook and the task's name is ending_hook?

You can also run capistrano with --trace to see the exact stuff that was run.

[vpillinger]

No, "rending_hook" was bad copy and paste. I changed the original hook names for confidentially and the original name started with an "r"

The failure hook was working when the deploy failed because of an interrupt, like ctrl+C, but not if I let it get to where it would fail via deploy:check

[mattbrictson]

Interesting! Looking at the code, that is indeed the behavior. That is, when a deploy fails due to a linked file not existing, the deploy:failed task is not invoked.

The exit 1 means that Capistrano exits immediately without any opportunity for failure tasks to be run.

capistrano/lib/capistrano/tasks/deploy.rake

Lines 90 to 93 in 220db8f

	unless test "[ -f #{file} ]"
	error t(:linked_file_does_not_exist, file: file, host: host)
	exit 1
	end

This behavior has been in place ever since Capistrano 3.0 was released.

I don't know if this behavior was intentional when it was written 5 years ago, or if this is a bug. Even if it is a bug (and I am inclined to agree it is) I am hesitant to change this behavior since it has been this way for 5 years and this is the first time the problem has ever been reported, to my knowledge.

The workaround would be for you to redefine the deploy:linked_files task to replace the exit 1 line with something like raise StandardError instead.

@will-in-wi and @leehambley is this something we should fix? I guess the risk is that there users are relying on the current behavior, because it would be a breaking change for them.

[will-in-wi]

I'd want to do some investigating in regards to what happens if we remove the exit (or move it elsewhere). I think in this case we do want it to exit with a status code of 1, but it'd be ideal if we could finish running the tasks. In short, I think this could be fixed/improved. I'll take a look at it.