PerlStalker's SysAdmin Notes

Testing for "Bitness" in Configuration Manager 2012 App Deployments

2012-05-01T14:27:30Z

We started our deployment of System Center Configuration Manager 2012 last week and I ran into an interesting problem.

One of the first apps I rolled out to test with was Strawberry Perl. I grabbed the 64-bit MSI and ran through the Create Application wizard and added the MSI to the deployment types. One quick deployment later and ConfigMgr was happily installing perl on my servers.

… Most of my servers.

You see, I still have a couple of 32-bit servers hanging around and the 64-bit MSI wouldn't install. D'oh. So, I figured it would be easy to jump into the Requirements of the distribution and limit the package to 64-bit systems. It wasn't. While there are options for RAM amounts, CPU speed and disk space there's nothing to test for CPU architecture.

To fix this, I created a new Global Condition to test for the "bitness" of a server.

The information I'm looking for is in the AddressWidth property of the Win32_Processor class. You can see the list of properties by running gwmi Win32_Processor in powershell. If you run

gwmi -query "select * from Win32_Processor where AddressWidth = 32"

and get back a screen full of text, your system is 32 bit. If you specify the wrong value for AddressWidth, the command will exit with no output.

The condition properties should look something like this when you're done.

Once that's done, it was a simple matter of adding the check to the deployment.

Set the Value field to 64 for 64-bit systems and 32 for 32-bit systems and you're done.

Having said all that, I'm still not sure that I haven't missed a setting somewhere. One would think that a test to see if an app matches the target architecture would have been a no-brainer to include. If there's a setting I missed, please let me know because not having it just doesn't make any sense.

Update 2012-05-01 Tue 08:25: One of my co-workers pointed out that there is, in fact, an "easier" way.

When you add a requirement, there's an option for Operating system. In the tree view at the bottom of the pane, you can select just the 64-bit version of the OS that you're targeting.

I totally missed that before. It takes a bit more clicky-clicky to use for every 64-bit or 32-bit only package you deploy but it may be more obvious to the next admin.

gPodder and EMMS

2012-04-30T16:32:37Z

I use gPodder to pull podcasts for all my listening pleasure at work. I had been using gnome-mplayer to listen to them but after a recent re-install, gnome-mplayer started to hanging every when I pause the playback. Gmplayer works fine so I know that mplayer, by itself, is not the problem but that does have a systray icon to make it easy to click and pause playback.

Since I'm so frequently using emacs, I thought to myself "I wonder if I can do this in emacs?" Of course, I can. Enter emms.

emms Configuration

Based on a little bit of research, I decided to use mplayer as the backend. The setup is pretty easy. This is what I'm using based on this blog post.

;; emms

(require 'emms-player-mplayer)
(require 'emms-source-file)
(require 'emms-source-playlist)

(setq emms-player-mplayer-command-name "mplayer"
      emms-player-mplayer-parameters '("-slave")
      emms-player-mpg321-command-name "mpg123"
      emms-player-list
      '(emms-player-mplayer
        emms-player-mplayer-playlist
        emms-player-mpg321
        emms-player-ogg123))

gPodder Configuration

Now that I can listen to the music via emacs, I need to tell gPodder how to feed the podcasts into emms. It's simple to add them to playlist from outside of emacs with this:

emacsclient --server-file=work --eval '(emms-add-file "TD140.mp3")'

In $HOME/.config/gpodder/gpodder.conf, you can set the player setting. gPodder replaces the %U with the file name. … In theory. In practice, gPodder only replaces %U if it's at the end of the line. That means I needed to write a quick wrapper.

CANNOT INCLUDE FILE upload/add2emms.pl

I run emacs in daemon mode (with TCP). If you don't, you'll want to change the $emacs variable.

And you're done. Happy listening.

IkiWiki and Org-mode

2012-03-26T04:02:39Z

I love emacs. I practically live in emacs at work. One of my favorite features is org-mode. At its simplest, org-mode is simply a markup language similar to wiki-text or markdown. When you get into it more, the real power of scheduling and task management comes out. In my case, it's the only system that's I've been able to use to stay organized. I've been so happy with org-mode that I start to looking into using for this blog.

I looked at several options from using org-mode's publish feature to publish HTML that I could paste into Drupal or using the native publish feature on its own. In the end IkiWiki won out.

IkiWiki has several useful features for me; mainly, the RSS/Atom feeds and a tag cloud. Most important, the formatting language is pluggable and there's a plugin to use org-mode instead of the native markdown. Getting it setup was pretty straight forward but had a couple of hang-ups because of my personal setup and there are a couple of things to look out for when actually using it.

Installing

The first step (after installing IkiWiki) was to get Chris Gray's excellent new_org plugin. This is what does the translation of the org files to HTML. Actually, it doesn't do the translation. Instead if feeds the file to emacs to spit out the HTML. That's were my first problem hit.

As I said, I already use emacs and one of the things I did on my server was start emacs in server mode. Once in server mode, I use emacsclient as my editor which really fast. The problem is that new_org also uses emacs server to keep from starting emacs for every change which speed up the translations. Unfortunately, the two servers conflicted. I think there was an issue with the two trying to use the same unix socket or something. I eventually switched emacs to use tcp sockets by putting (setq server-use-tcp t) in $HOME/.emacs.

There are also a few changes in new_org.pm needed to support tcp sockets. First emacsclient -s org-ikiwiki-compiler needs to be changed to emacsclient --server-file=org-ikiwiki-compiler. Second emacs --daemon needs to be changed to emacs --daemon=org-ikiwiki-compiler. There are a couple of places where those changes needed to be made. Once those were in place, I was almost able to publish files.

I ran into a problem with emacs hanging when I tried to publish with ikiwiki --refresh. It turns out that I was still using the ancient version of org-mode that came with emacs on Ubuntu Lucid. My problems went away after upgrading org-mode to the latest version.

Those items done, I was able to publish new posts all day long.

Publishing

Blog System Files

The first thing to keep in mind is that the index, sidebar and tag files all need to remain in markdown with .mdwn extensions. When I tried to change them to org-mode, IkiWiki would not process them correctly. That's a little annoying but no big deal, really, since I hardly ever need to touch them. I mostly org-mode for the blog posts.

IkiWiki Directives

This is where things got a little weird. IkiWiki directives use the same syntax as org-mode does for links. For the most part, the two play well together. For example, the tag directive does, in fact, add tags to the post. I did run into an issue with images, however. It's possible to use the basic org-mode link syntax to add an image but it doesn't properly deal with paths. IkiWiki's img directive takes care of that but emacs would see the HTML that the directive wrote and happily escape the HTML.

The solution is to wrap the img directive in #+BEGIN_HTML and #+END_HTML blocks and everything is happy.

I haven't used any of the other directives yet so I can't tell you how well they work. Though that may be why the index and tag pages where giving me grief.

Conclusion

With those little snags out of the way, everything is full steam ahead. I'll keep y'all informed of anything else I run into but it's been smooth sailing so far.

Renovations

2012-03-25T22:52:26Z

I've been a busy PerlStalker this week. As you may have noticed, this site has changed somewhat. I've recently decided to drop Drupal and switch to IkiWiki. Drupal has been good to me but it's really too heavy for my little blog. I have no use for most of the features that Drupal provides and it's not worth it to me to keep it up-to-date.

IkiWiki has a couple of really nice things going for it. First, it spits out static pages. Except for search and comments, there's no need for any dynamic content on my site. Google and Disqus fill both needs. I could use the IkiWiki CGI but there's really no need.

Second, I can compose my posts in emacs org-mode. It beats the heck out of writing all of my posts in straight HTML. I had to jump through a few hoops to use org-mode but it's working great now. I'll post more on that later.

By default, IkiWiki uses a pretty nice syntax called markdown. I chose not to use it because org-mode is, in many ways, simpler and more powerful than markdown and I use it for all of my note taking and documentation at work.

Now, a little housekeeping. One of the drawbacks to moving to IkiWiki was that I was unable to get it to use the original posting dates from Drupal. All of the imported posts have a line at the top that has the original posting date.

Along those lines, you've probably already seen the slew of reposts in your feed reader. That's a symptom of the conversion to IkiWiki. This should be the only time it'll be a problem.

Thanks for reading.

Setting the Default Monospace font in XFCE

2012-03-23T04:59:43Z

Originally posted at 2011-12-19 Mon 09:38

I use XFCE4 on my box at work because it's lightweight and still provide the features I want. I wanted to change the default font so that emacs used the font I wanted without having to change my .emacs file. Unfortunately, XFCE4 only lets you set system default but not the monospace font. Fortunately, XFCE4 uses fontconfig. All you need to do is edit $HOME/.fonts.conf and add this little XML snippet. (See the fonts.conf manpage.)

<match target="pattern">
        <test qual="any" name="family">
                <string>monospace</string>
        </test>
        <edit name="family" mode="assign">
                <string>Terminus-9</string>
        </edit>
</match>

If you have to create $HOME/.fonts.conf, make sure you wrap that snippet in <fontconfig> tags.

Running Emacs Functions in Batch Mode

2012-03-23T04:52:15Z

Originally posted at 2011-11-10 Thu 10:22

I use emacs for a lot of things at work. One of the more useful is org-mode for to do lists, scheduling and meeting notes. Org-mode can sync to mobile devices running the app MobileOrg. Unfortunately, that sync is a manual process. The good news, emacs is scriptable and can be run in batch mode to automate things. Here are a couple of things I use.

Note: Emacs batch mode spews a huge amount of crap to stderr. If you're putting these in cron, you may want to slap a 2>/dev/null on there to keep it quiet.

Sync with MobileOrg

The first line pushes the local org files to sync and the second pulls changes coming from the device.

emacs --batch -l $HOME/.emacs -f org-mobile-push
emacs --batch -l $HOME/.emacs -f org-mobile-pull

Publish Meeting Minutes

I keep meeting minutes using org-mode and then publish them to my web site. I also maintain my personal work site using org-mode. To publish all configured sites, you can use the script below.

emacs --batch -l $HOME/.emacs -f org-publish-all

If you want to publish a specific site rather than all of them, you can use this incantation.

emacs --batch -l $HOME/.emacs --eval '(org-publish "meetings")'

Updated 2012-03-22 Thu 22:49: It's also possible to use --eval with emacsclient. You, of course, need emacs server running but it makes those batch runs much faster.

Finding Old Computer Accounts in Active Directory with Powershell

2012-03-23T04:44:47Z

Originally posted at 2011-06-20 Mon 13:31

Following up on a previous post discussing finding old user accounts in Active Directory, here's how you find old computer accounts.

This works on basically the same premise as the user script. In short, we're going to check the last time the computer logged into Active Directory. That happens on every reboot and from time to time while the machine is up. The same warning applies to computers as it does for user accounts. The last logon timestamp is only accurate to about a week. Since people are generally checking back six months or more, it isn't much of an issue.

To get the list of stale machines:

get-adcomputer -properties lastLogonDate -filter * | where { $_.lastLogonDate
-lt (get-date).addmonths(-12) } | FT Name,LastLogonDate

You can throw a sort in there if you'd like.

get-adcomputer -properties lastLogonDate -filter * | where { $_.lastLogonDate
-lt (get-date).addmonths(-12) } | sort Name | FT Name,LastLogonDate

It's just as easy to delete all of those accounts. I've added a -whatif to make it harder to do something stupid with cut-and-paste. Take it off to actually delete the computer accounts.

get-adcomputer -properties lastLogonDate -filter * | where { $_.lastLogonDate
-lt (get-date).addmonths(-12) } | Remove-ADComputer -whatif

Updated 2011-06-20 14:15: Added a missing } to every example. Don't you hate it when you have a bug in your code and you copy and paste that same line over and over again? Me, too.

Convert a Moose Object into a Subtype

2012-03-23T04:38:53Z

Originally posted at 2011-06-03 Fri 14:01

Have you ever had the situation where you had a parent class but you find out later in the code that you really want a subclass? With perl Moose, it's really easy to do.

Consider the classic point example described in Moose::Cookbook::Basics::Recipe1. Suppose I start with a Point but I decide later that I need that point to shift into 3D. I don't want to change the values for x and y, I just want to add the new z point. I could brute force it by creating a new point and copying the values over from the old point but that feels like more work than it should be. Well, it is.

Fortunately, Moose uses Class::MOP::Class which provides the rebless_instance method which can convert a object into a subtype.

package Point;
use Moose;

has 'x' => (isa => 'Int', is => 'rw');
has 'y' => (isa => 'Int', is => 'rw');

package Point3D;
use Moose;
extends 'Point';

has 'z' => (isa => 'Int', is => 'rw');

package main;
use Data::Dumper;

my $p1  = Point->new(x => 4, y => 3);
print Dumper $p1;

$p1 = Point3D->meta->rebless_instance($p1);
print Dumper $p1;

And the output:

$VAR1 = bless( {
                 'y' => 3,
                 'x' => 4
               }, 'Point' );
$VAR1 = bless( {
                 'y' => 3,
                 'x' => 4
               }, 'Point3D' );

Now $p1 is a Point3D object rather than a Point object.

You can go backwards, too, with the rebless_instance_back method but you face the potential for data loss when you do that.

JSON Serialization with Dancer and MooseX::Storage

2012-03-23T04:31:34Z

Originally posted at 2011-04-27 Wed 16:07

I'm hacking on a new tool in perl to manage my KVM cluster. Part of it is a RESTful interface using JSON. The objects I'm using are written using Moose and MooseX::Storable to simplify serialization. I can convert objects back and forth between perl objects and JSON all day. Unfortunately, there's a fly in the ointment.

I'm using Dancer to provide the framework for the RESTful interface. The nice thing about Dancer is that it can automatically serialize perl data structures and it can do it via JSON. That's all shiny except Dancer's serializer doesn't know about MooseX::Storable. In fact, Dancer chokes when I try to send an object back from one of the handlers.

All is not lost. There is a work around that appears to work in my tests. First of all, I needed to tell Dancer that it's okay to try and convert my objects. This is done in config.yml.

engines:
  JSON:
    convert_blessed: '1'

That tells Dancer, via the JSON package, to try to convert objects using the method TO_JSON. But, wait! I don't have one of those. Yet. Fortunately, MooseX::Storable has everything I need.

The MooseX::Storable API offers a couple of options. The first is to use freeze() to turn the object into a JSON string. That works if I'm only returning a single object but I'm often returning lists of things which complicates things.

sub TO_JSON { $_[0]->freeze }

Actually, it seems to work but Dancer has goes through the returned JSON string and encodes before sending it on. That's actually a second round of encoding since MooseX::Storable did it once already. Here's what a list containing an object looks like.

[
   "{\"__CLASS__\":\"YAVMM::Host-0.01\",\"cluster\":\"dev\",\"name\":\
"rbsmith-desktop\",\"ram\":3924056,\"num_cpus\":4}"
]

Ugly, huh? In fact, that might screw up anything non-perl that tries to use the RESTful interface. We can get around that by going down one level and using pack() instead.

pack() and the corresponding unpack() are required by anything implementing the MooseX::Storage API. Rather than going all the way to JSON, pack() spits back a serialized perl hashref. It turns out that that is exactly what Dancer needs to properly encode the object. So, I changed TO_JSON to look like this:

sub TO_JSON { $_[0]->pack }

The new output is below. Isn't it pretty? That's exactly what the JSON should look like.

[
   {
      "__CLASS__" : "YAVMM::Host-0.01",
      "cluster" : "dev",
      "name" : "rbsmith-desktop",
      "ram" : 3924056,
      "num_cpus" : 4
   }
]

On the client side, I still need to decode that mess. That's actually very easy using the JSON module and feeding the JSON text into decode_json(). That spits back a perl hash or list reference that has the (mostly) deserialized data. I say mostly because my objects are still hashes and not objects. All that I need to do is feed those hashrefs to unpack() and I get my object back.

Here's how I would translate that list above. Note: My object is called YAVMM::Host. You would, of course, use your own class name instead. Also, assume that $json_text has the JSON text. How it got there is beyond the scope of this document.

my @hosts = map { YAVMM::Host->unpack($_) } @{ decode_json($json_text) };

Now @hosts is a list with a bunch of YAVMM::Host objects that I can have all sorts of fun with.

Ubuntu Fails to Boot with Spice Console

2012-03-23T04:31:34Z

Originally posted at 2011-04-12 Tue 15:44

I needed to roll a couple of debs for my Ubuntu servers at work. Since I run Gentoo on my desktop, I rolled a 64-bit Lucid VM. Unfortunately, after the install finished, I got the following error on the console and the VM wouldn't boot.

error: no suitable mode found
error: unknown command 'terminal'

I figured the problem had something to do with the interaction with spice console (which I'm using instead of VNC). Fortunately, the fix is pretty easy.

Boot the install CD and go into Recovery mode.
Drop into a shell on the system.
Make sure /boot is mounted in case it's on a separate partition.
Edit /etc/default/grub and set GRUB_TERMINAL=console
Run /usr/sbin/grub-mkconfig > /boot/grub/grub.cfg to update /boot/grub/grub.cfg
Exit the recovery shell and reboot

Poof! The VM now boots and I can log in.