Collated multiline output from knife ssh

Branch CHEF-2395.

Pull requested: https://github.com/opscode/chef/pull/97

Marking resolved as there is a patch attached.

There are buffering consequences to this that need to be resolved.

1) This prevents you from monitoring servers in real time. Consider these use cases:
knife ssh role:my_servers 'vmstat 2'
knife ssh role:www 'tail -F /var/log/apache2/access.log'

2) Because each server is buffered until the command is complete, if something occurs to prevent the command from completing you never see why. At the very least, an INT should print the buffer.

I'm not sure how to rewrite this to fill these use cases, but I think these are current use cases and take priority.

For 1) I'd say, just have an option so you can active/deactivate buffering as desired. This is similar to an existing option in onall (http://code.ticketmaster.com/index.php?page=onall-usage).

For 2) this is because there isn't a timeout (is there?) for commands because you might be doing something that is long running. Can we either catch ctrl-C or some other exit command that would cause buffers to be flushed before terminating any running commands? You wouldn't see the error in realtime, but you would at least be able to see it after bailing and then you could try again with buffering turned off if needed.

Any news/progress on this topic? Has it been superseded by other functionality?

I think it's still a viable feature if someone wants to code it up with regard to the mentioned constraints.

What about some "aggregated" outputs à la dsh?

If you run a command on N nodes and most of the output is identical, knife ssh should aggregate everything and report which node did things differently.