made a modified [/usr/bin/ruby193-rake] that included output of the path and 'which rake':

#!/bin/bash

COMMAND="echo $PATH && which rake &&  rake $@"
scl enable ruby193 "$COMMAND"

when ran from normal shell you see:

# /usr/bin/ruby193-rake
/usr/local/sbin:/usr/local/bin:/sbin:/bin:/usr/sbin:/usr/bin:/root/bin
/opt/rh/ruby193/root/usr/bin/rake
rake aborted!

I'm not familiar enough with how SCL works to know how it includes commands that are not in the shell's PATH to be found as an executable but when ran from within puppet with a simple manifest:

 exec {"test_rake":
    cwd         => "/tmp",
    environment => ["HOME=/tmp"],
    command     => "/usr/bin/ruby193-rake",
    path        => "/sbin:/usr/sbin:/bin:/usr/bin",
    user        => "root";
  }

you get:

# ruby193-puppet apply repro.pp 
Notice: /Stage[main]//Exec[test_rake]/returns: /sbin:/usr/sbin:/bin:/usr/bin
Notice: /Stage[main]//Exec[test_rake]/returns: which: no rake in (/sbin:/usr/sbin:/bin:/usr/bin)
Error: /usr/bin/ruby193-rake --help returned 1 instead of one of [0]
Error: /Stage[main]//Exec[test_rake]/returns: change from notrun to 0 failed: /usr/bin/ruby193-rake --help returned 1 instead of one of [0]
Notice: Finished catalog run in 0.55 seconds

but if ran *without* the path =? ...  it runs rake just fine:

# ruby193-puppet apply repro.pp 
Notice: /Stage[main]//Exec[test_rake]/returns: executed successfully
Notice: Finished catalog run in 0.81 seconds
[root@sat6-mdp1 ~]# vim repro.pp 
[root@sat6-mdp1 ~]# ruby193-puppet apply repro.pp 
Notice: /Stage[main]//Exec[test_rake]/returns: /opt/rh/ruby193/root/usr/bin:/usr/local/sbin:/usr/local/bin:/sbin:/bin:/usr/sbin:/usr/bin:/root/bin
Notice: /Stage[main]//Exec[test_rake]/returns: /opt/rh/ruby193/root/usr/bin/rake
Notice: /Stage[main]//Exec[test_rake]/returns: rake aborted!
Notice: /Stage[main]//Exec[test_rake]/returns: No Rakefile found (looking for: rakefile, Rakefile, rakefile.rb, Rakefile.rb)
Notice: /Stage[main]//Exec[test_rake]/returns: 
Notice: /Stage[main]//Exec[test_rake]/returns: (See full trace by running task with --trace)
Error: /usr/bin/ruby193-rake returned 1 instead of one of [0]
Error: /Stage[main]//Exec[test_rake]/returns: change from notrun to 0 failed: /usr/bin/ruby193-rake returned 1 instead of one of [0]
Notice: Finished catalog run in 0.75 seconds

https://github.com/Katello/katello-installer/pull/38

I've narrowed this down to the update of scl-utils from scl-utils-20120927-2.el6.x86_64 to scl-utils-20120927-2.el6_4.6.x86_64.  Before updating my host from one to the other, I couldn't reproduce it using the example in comment #1, now I can.  I'm pretty sure this is due to the change introduced in bug #976551.

The scl tool now checks to see if the SCL is already enabled by checking the $SCLS environment variable, then doesn't run the enable script to export $PATH etc if it was already enabled.  Since we're running Puppet under the SCL (via ruby193-puppet), $SCLS is already "ruby193".  By specifying "path" in Puppet, it exports PATH and since $SCLS still contains "ruby193", the scl tool doesn't re-enable/export $PATH for the collection.

So yes, I think your fix is correct Mike, we shouldn't override $PATH.

QE test notes:

Install latest and greatest RHEL 6.4, then install/configure katello*

Both repo and ISO will need to be tested.

Additional note:  this involves foreman so of course install katello-foreman-all, etc.

HOTFIX:

If you wish to deploy a fix ASAP do the following:

1) Download hotfix RPMS:

# wget http://mmccune.fedorapeople.org/hotfix/994326/katello-configure-1.4.3-17.el6sat.noarch.rpm
# wget http://mmccune.fedorapeople.org/hotfix/994326/katello-configure-foreman-1.4.3-17.el6sat.noarch.rpm


2) Upgrade RPMs on your Satellite 6 host:

# rpm -Uvh katello-configure-1.4.3-17.el6sat.noarch.rpm katello-configure-foreman-1.4.3-17.el6sat.noarch.rpm

3) proceed with katello-configure as indicated in the Installation Guide

Verified in install from repo (cswiii) and from ISO (gkhachik). With latest RHEL packages (and specifically scl-utils), installer does not fail, and no rake "command not found" errors occur in log(s). 

QE Verified.

This was verified and delivered with MDP2. Closing it out.

This was delivered and verified with MDP2. Closing the bug.