unicorn service can't be restarted and new routes are not found after deployment

[benams]

My server is ubuntu 14.04. I can't see the unicorn service in /etc/init.d, and I can't service unicorn restart or service unicorn_myapp restart. Otherwise - the process id remains the same as it was, while the workers' process ids are changed ( I don't really know what it means)

The only way to restart and get a new process id is to kill -9 <process_id> and then to cap production deploy again.
If I don't do so, each time I add a new route it just can't be found...

1. Do you know why the route is not found?
2. Does it make sense that I can't restart the service?

Here is my capistrano deploy script:

set :application, 'myapp'
set :repo_url, 'git@10.0.0.101:/opt/git/myapp.git'

ask :branch, proc { `git rev-parse --abbrev-ref HEAD`.chomp }.call

set :use_sudo, false
set :bundle_binstubs, nil
# set :linked_files, fetch(:linked_files, []).push('config/mongoid.yml')
set :linked_dirs, fetch(:linked_dirs, []).push('log', 'tmp/pids', 'tmp/cache', 'tmp/sockets', 'vendor/bundle', 'public/system')

after 'deploy:publishing', 'deploy:restart'

namespace :deploy do
  task :restart do
    invoke 'unicorn:reload'
  end
end

[kunerd]

I have a similar issue. Does calling cap production unicorn:stop and then cap production unicorn:start is working for you?

[kunerd]

I think the order of stopping and starting is wrong in restart and reload. If I use:

  task :restart do
    after 'unicorn:stop', 'unicorn:start'
    invoke 'unicorn:stop'
  end

it works. i think the problem is that start is invoked while the old version is still running. Therefore the if statement in start prevents the call to start unicorn.

@benams A quick fix for your reload should be:

task :restart
    on roles(fetch(:unicorn_roles)) do
      within current_path do
        info "reloading..."
        execute :kill, "-s HUP", pid
      end
    end
    invoke "unicorn:start"
end

Maybe you need an after hook like in my solution above.

[ahorner]

@benams: Your deploy script invokes unicorn:reload, which doesn't actually bring up the new application code in cases where the unicorn configuration has preload_app: true. You'll want unicorn:restart in most cases. See https://github.com/tablexi/capistrano3-unicorn/blob/master/examples/unicorn.rb for a sample zero-downtime deploy configuration.

Unicorn doesn't have any sort of default service script. You can manually add one yourself (something like https://gist.github.com/jaygooby/504875 might do the trick).

@kunerd: The invoke "unicorn:start" lines are there to ensure that a process is actually available to receive the relevant restart signal, not to start a new unicorn process after killing the old process. unicorn handles the restart/forking of new processes in response to receiving a HUP or USR2 signal. You should ensure that you're using the correct task for your configuration.

[josecoelho]

No success using unicorn:restart

The signal USR2 is sent to the unicorn service, but the next QUIT signal isn't being executed. unicorn SIGNALS docs

My preload_app is false.

Solve the problem using unicorn:stop, then unicorn:start.

I'm not looking for zero-downtime deploy.

Hope this helps.

[FrankFang]

unicorn:stop, then unicorn:start.
It works for me.

[crice88]

I'm running into this issue intermittently. Depends on what code is changed, having a hard time pinning it down though.