might we want per connection values?

:except => {:no_release => true}

It is common to configure tasks to 'announce' deployments in IRC, Campfire, etc. If you have 6 app servers, you don't want to see 6 announcements. In Capistrano, this is handled via the :no_release => true flag. Various tasks only execute on the 'release' servers.

An easier way to meet this would be to introduce a :release role in the default setup

role :release, "app1.example.com"

remote_task :announce_in_irc, :roles => :release ...

Drawback: Yet another thing to change when you migrate a project from cap to vlad

'dynamic deployments'

role :app, "app1.example.com"
role :app, "app2.example.com"

Let's say that app1 and app2 need slightly different monit configurations.

In Capistrano, you might approach this by making two additional roles, and splitting your 'push a monit config' task into two. This sucks.

Vlad makes the 'execution context' of a task available. In Vlad, you would:

remote_task :update_monit, :roles => :app
  rsync "templates/#{target_host}.monitrc", "/etc/monitrc"
end

fine-grained tasks

remote_task :update
remote_task :symlink
remote_task :migrate
remote_task :deploy => [:update, :symlink, :migrate, :restart]

Let's assume that this is a multi-server config with shared deploy path. The user wants to do only a single checkout. If we make “update” be one big task body that includes the update, symlink, and migrate steps, it is difficult for the user to override the roles for the particular steps they need to change.

If we break these into separate tasks, they can say:

Rake::Task["migrate"].options[:roles] = :master_db

and the migrations will only run on the master db

sudo / via how? and if we call it via I will stab ppl. “user” is sufficient.

handling 'use_sudo'

1. Check for this inside the 'run' command, and preface the command with 'sudo' if necessary

2. Default this to 'false' in the reset method, and check for it in the default tasks that we provide:

   if use_sudo then
     sudo "blah"
   else
     run "blah"
   end
   

Option 2 has fewer moving parts, but clutters up the tasks that care about this.

Dependencies

Task dependencies aren't settable when creating a Rake::RemoteTask.

Apache configuration

Pull in railsmachine/rails/recipes/apache.rb's apache configuration. Needs erb to work.

I really like tasks with naming <cmd>_<role> (eg setup_app, start_web). We could easily make the front end remote_task command look for such a convention and apply the :role => x automatically.

from bousquet: get a couple of server environment recipes that prepare your machine that would be the golden ticket:

rake vlad:prepare TYPE=accelerator | ubuntu | osx | osxserver | site5 | ...

and have people maintaining those setups who depend on them