Too bad that the semester is nearly over now. Oh well. I need to finish up my documentation for how a newbie can create a Eucalyptus cloud. I was really hoping to get into some source code before the semester was over but that is just not going to happen. Even though I haven’t written a line of code this semester, the course has been invaluable. I am now far more comfortable interacting with software developers via IRC. I have learned a good deal about Linux system administration through failing over and over again. I still hate writing documentation, but now I have gotten a little practice at least.

Hopefully I can wrap up my documentation project soon and get working on the final assignments for the course: an ethical analysis of Eucalyptus and a slideshow presentation about the work I did over the semester.

In other news, the WSU programming team (of which I am a part of) got 4th place out of 27 at the competition held at Quinnipiac University! We solved 4 (nearly 5) out of the 6 problems. I wasn’t expecting such a high so it came as a pleasant surprise to see that we got 4th. This was my last year on the team so I’m glad we performed well for my final competition.

Oh, before I forget: after this semester is over… I’m done with undergrad! Which means I need to find work, preferably developing FOSS. Just need to finish my resume. So, who wants to offer me a job?

After lunch, Andy and I spent the next 3 hours trying to resolve the various issues that were keeping our cloud from functioning properly. The good news is that I learned a lot about how Eucalyptus works, networking, and some neat command line tricks. The bad news is that our cloud is still not functioning.

As I pointed out in my last post: networking is hard. It is especially hard when working on the university network whose topology is pretty much unknown to me. The first big mistake that I made when I set up the network for the cloud was split the LAN into 2 subnets. The CLC, SC, and CC were on one subnet, while all the NCs were on their own subnet and the CC acted as a router for them. I had misread some networking documentation and mistakenly thought that having the NCs are on their own subnet would make things easier for later on. Andy set me straight about that and I “flattened” the network out so all of the NCs are now on the same subnet as the other Eucalyptus components. Next, we set up ethernet bridges for virtual networks that will be needed for the virtual machines. That was mostly straightforward but through some slight misconfiguration I encountered Red Hat networking hell involving a renamed `eth0` which Andy resolved for me by messing with udev.

I thought that I had a recent enough kernel installed on all of the machines to avoid the kernel bug that breaks Walrus. Turns out I didn’t. So no matter what we did the cloud would never function until that was resolved. A `yum update kernel` and a reboot took care of that. Then, for some strange reason, PostgreSQL processes wouldn’t start. Andy started them manually and things seemed to work just fine for subsequent reboots.

Despite fixing all of these issues and making a ton of progress in such a short time, we were still unable to run a VM instance. Andy and I ran out of time (he had a train to catch and I had to return to work) but he suggested that I essentially start over once again. This time I will be using the Eucalyptus packages that are available from the Eucalyptus repository.

So, another week goes by without a working cloud… but I learned a lot. Many thanks to Andy Grimm for making me a better system administrator.

I was using the “managed” networking mode for Eucalyptus. However, upon testing if the network was “VLAN clean.” I followed the instructions here and alas, the VLAN was not clean. It seems like more of a headache to get the VLAN to function than it is to switch to the “managed-novlan” networking mode.

The “managed-novlan” mode required me to manually configure an ethernet bridge for Eucalyptus to use. I followed the instructions in the Eucalyptus 3.0 Installation Guide and everything seems to work (running `service network restart` shows no errors), but now the Cluster Controller cannot communicate with the Node Controllers. This means that there are no available resources advertised for VMs. So now I cannot even try to start up a VM instance.

I only have basic networking knowledge so these issues have been a real headache for me. The fine folks on the #eucalyptus IRC channel have been helping me out with everything but everytime we eliminate an issue there is another one to take its place.

In other news, Eucalyptus project member Andy Grimm will be visiting WSU this week. I’m looking forward to talking with him and hopefully I can convince him to give me some on-site assistance with the servers.

Prerequisites

This guide makes the assumption that you already have a functional web server (Apache, nginx, etc.), PHP 5, and database server (either MySQL or PostgreSQL).

Setup Database

I made a MySQL user specifically for GNU FM

CREATE USER 'gnufm'@'localhost' IDENITIFIED BY 'yourpassword';

Create a database

CREATE DATABASE gnufm;

Download Source

GNU FM uses Git for version control. The source is available from gitorious. Clone the repository with the following command.

git clone git://gitorious.org/foocorp/gnu-fm.git

Install Gnukebox

From the gnu-fm directory, copy the server application, gnukebox, to the web server directory.

cp -r gnukebox/ /var/www/

From your web browser, go to http://yourwebserver/gnukebox/install.php. It’s quite straight forward. Enter the information in for your database server like the database name and the user that you have created.

If the web server cannot write to the directory in which you installed gnukebox, you will see a page that prints out the configuration code that you will need to manually insert into the config.php file in the gnukebox directory.

Install Nixtape

From the gnu-fm directory, copy the web interface, nixtape, to the web server directory.

cp -r nixtape/ /var/www/

From your web browser, go to http://yourwebserver/nixtape/install.php. You will enter the same database details as you did for gnukebox and give your GNU FM server a name.

Once again, if the web server cannot write to the directory in which you installed nixtape, you will see a page that prints out the configuration code that you will need to manually insert into the config.php file in the nixtape directory.

Nixtape requires some extra PHP libraries to be installed along with it. Please note that the version numbers of these libraries may be different depending on how far you in the future you are coming from. Which means that copy/paste’ing the commands I give you will not work, but you’ll probably only have to change the version number in the command to make it work.

Install ADOdb

ADOdb is a database abstraction library for PHP. Installing the library takes just a simple apt-get.

apt-get install php5-adodb libphp-adodb

Install Smarty

Smarty is a PHP templating library that nixtape uses for all web pages. There are packages available in the Debian Squeeze repository, but GNU FM does not work with them. The problem is the result of a slightly different path to the files contained in the Debian package. Since the GNU FM repo has a .tar.gz archive of Smarty within it, we’ll just install that instead.

From the gnu-fm directory, cd to the external_dependencies directory.

Extract Smarty and the Smarty text plugin.

tar zxf Smarty-2.6.22.tar.gz
tar zxf smarty-gettext-1.0b1.tgz

Copy the needed files to the nixtape directory.

mkdir /var/www/nixtape/smarty
cp -r Smarty-2.6.22/libs/ /var/www/nixtape/smarty
cp smarty-gettext-1.0b1/block.t.php /var/www/nixtape/smarty/plugins

Create a templates_c directory for compiled templates and make it writeable by the web server user.

mkdir /var/www/nixtape/themes/gnufm/templates_c
chgrp www-data /var/www/nixtape/themes/gnufm/templates_c
chmod 775 /var/www/nixtape/themes/gnufm/templates_c

Profit!

At this point I was able to have a working GNU FM installation. There are still plenty of things left to do such as add music, add a new theme, etc. Maybe once I get around to doing those things I will make a follow-up tutorial. Anywho, I hope this has been of some use to you folks out there in Internet land. Still having a hard time setting up GNU FM? If so, head on over to #libre.fm on irc.freenode.net and ask the developers for help. Oh, and don’t forget to leave a comment if there’s an error or something you would like me to improve in this tutorial. You can also leave a comment and say what a great guy I am.

Take it easy.

On Friday, Prof. Wurst and I set up several of the computers and installed CentOS 6.2 on them. I have already read and written a lot of documentation about how to compile Eucalyptus from source, so it did not take me very long to get the Eucalyptus services running on the new machines. The services, aside from the NCs, are running for the most part on 3 machines now. We decided that it would be best to let the Walrus service have its own server where previously it was on the same server as the CLC. So, the CLC, Walrus, SC, and CC are operational again. I have registered the Walrus, SC, and CC servers (CC and SC are on the same machine, by the way) with the CLC. However, the system is reporting that the SC and Walrus are broken and I have not yet resolved that issue.

In the meantime I decided to set up a couple of NCs, but in trying to set up the ethernet bridge for KVM to use, I managed to lock myself out of one of the servers when I did a `system network restart`. I guess I screwed up the configuration files! It’s not much of an issue because tomorrow I will be going to campus to work in the server room with Mike Iudiciani and I will resolve the issue there.

In doing all of this work, I have found a bug in the `euca_conf` tool. One role of euca_conf is to get credentials for access to the Eucalyptus system via command line. The command line switch for this is –get-credentials. I started the `eucalyptus-cloud` service and tried to get my credentials via euca_conf. However, I just got a “list index out of range” error printed to my terminal. After waiting for some time, the command worked successfully and I got my credentials. The problem appears to be that the cloud services were not fully functional when I tried to get the credentials, and euca_conf did not properly handle the situation and crashed. I will be filing a bug report about this shortly.

In other news, spring break is this week, but since we are running way behind schedule I will be working on this project a lot over the break. Setting up these servers has proven to be much more difficult than anticipated so I will be spending a lot of time trying to catch up to where we should be and hopefully the result of all of this hard work will be a finally functional Eucalyptus cloud.

As an initial test, I’m using 3 of our servers to make a minimal working configuration. There are a lot of things that have to be done to each machine such as adding repositories, downloading dependencies, initializing PostgreSQL, and finally initializing Eucalyptus. There are 3 init scripts that Eucalyptus provides: eucalyptus-cloud, eucalyptus-cc, and eucalyptus-nc. Walrus, CLC, and SC require eucalyptus-cloud to be running. CC require eucalyptus-cc. Each node requires the eucalyptus-nc service. On the CLC machine, I registered it to be used for Walrus and the SC. We have another server, named Trinity, being used for the CC that needs to be registered with the CLC.

Now that the CLC and CC were operational, it was time to start up a node. This is where I ran into even more trouble. The computers we’re using as servers are older machines with CPUs that do not support hardware virtualization. With the release of RHEL 6, KVM is the supported hypervisor for virtualization. Unfortunately, KVM requires hardware virtualization support in order to function so it is not a possible option for us to use. The Xen hypervisor, on the other hand, doesn’t need to have hardware virtualization in order to function. The problem is that support for it was dropped in RHEL 6 so it is not so trivial to install now. The Xen wiki has a tutorial about installing Xen on a RHEL 6 system, which is quite long. I am slowly working through this tutorial to build the Xen packages that I need. It is an unfortunate situation because it is a very time consuming process. I was really hoping to have this system functional by today so I am a bit disappointed that I once again failed to meet my goals.

To end on a positive note, I am starting to become familiar with the Eucalyptus software. The more I start/restart the services and register/deregister components, the less confused I become about how the Eucalyptus system functions. I have written quite lengthy documentation on the CS401 wiki about all of my progress that another classmate has been using to help setup other machines that will later be used as nodes.

Now, how badly does Eucalyptus fail? By my calculations they had 55 points of fail which means that “Babies cry when your code is downloaded”. I think I may have been a bit too rough with my analysis. I really do like the Eucalyptus project, but as a new user that is also completely new to the cloud computing concept, it’s hard to figure out exactly what to do and how to do it. There is documentation but there is not enough of it. From my understanding the Eucalyptus team acknowledges that and one of our goals this semester is to improve documentation for users, and who better to write documentation for new users than… new users!

In addition to the documentation fails (I forgive you, Eucalyptus devs, I hate writing documentation as well , as far as I can tell there are no unit tests for the source code to check for bugs that may have creeped in. That alone earned Eucalyptus +25 points of fail and pushed the project into the crying babies category.

Once again, from what I have seen, the Eucalyptus project is great. Their developers have been supportive and responsive when I’ve had issues. As a FOSS advocate, the freedom of using a free and open source cloud computing platform far outweighs the convenience of a proprietary system that is easier for newcomers. Eucalyptus is much more win than fail, and I hope that my class can help the developers improve the usability of the software.

Understanding Eucalyptus

I came into this project knowing literally nothing about cloud computing. But from reading the Eucalyptus documentation, I have a vague understanding of what the various components do.

The first component is the Cloud Controller (CLC). The CLC is the gateway for access to a cluster. It basically runs the show. Note that I said clusters, plural. The CLC make high-level decisions and talks to the cluster controllers. Next, is Walrus, a persistent storage system for the users virtual machine images. The data is organized into buckets (get it?). Walrus can be installed on the same server as the CLC. The Cluster Controller (CC) is in charge of managing several Node Controllers (NCs). The CC also handles access to the Storage Controller (SC) which interfaces with different storage methods. The NCs are servers that run a virtual machine hypervisor (such as KVM or Xen).  The hypervisor manages all the virtual machines (VMs) that are running on the server. Put it all together and you have yourself a cloud!

Preparing Servers

My class has 8 machines running CentOS 6.1. One server will be the CLC, running Walrus as well. One server will be the CC/SC. The remaining 6 will be our nodes. If combining tasks on the servers becomes a problem, it shouldn’t be too difficult to move them to other servers, the only downside being that we will have less NCs.

I was confused about which VM hypervisor was going to be installed on the machines. All of the Eucalyptus 2.0 guides were giving instructions to install Xen. However, when I tried to install the Xen packages, they were not found. After some googling, I learned that with CentOS 6, support for Xen was dropped in favor of KVM.

Compiling from Source

I grabbed the latest and greatest code from the Eucalyptus Launchpad page. I found 3 helpful documents that allowed me to successfully build Eucalyptus. one two three. I’ve been documenting the process as I go on the CS 401 Wiki. It took A LOT of reading and some help from the #eucalyptus IRC channel, but I did it! CS 401 is now in possession of some fresh 3.1 binaries.

Installing Eucalyptus

Now that I have Eucalyptus 3.1 built, I need to install it on all of our machines. I have been using rsync to copy the binaries to the other servers. The CLC, CC, SC,  Walrus, and NC require different Eucalyptus services to be running in order for them to function. As of this writing I am currently having some trouble initializing the databases on the NCs.

I’ve made more progress this past week than any other week so far. Hopefully we can have a functional system in a week’s time.

What did we learn how to do? Well, Karl Wurst made us some test repositories on the CS git server. This included a few example files that we would edit. My team “cloned” the repository and made the changes, and when some of us tried to “push” the changes back… CONFLICT! Prof. Wurst had designed the exercise in such a way that we would see a merge conflict occur. When a merge conflict happens, git alters the conflicted files by putting in both your version and the conflicts marked by special characters. I simply deleted what was unnecessary to resolve the conflict, and “committed” the changes to my repository. Now I was able to push successfully.

In System Administrator land, Mike Iudiciani and I have been working to put together a small cluster on which we will install the Eucalyptus software. We’ve run into a lot of annoying problems with OS installation and the networking configuration. We were first plagued with bad install media, and then we had trouble getting the machines online. Once we finally had good install CDs, I talked to IT about how to configure our machines network interfaces and voila, we were online. Now begins the process of installing the Eucalyptus software.

The class voted on which project we would be working on for the semester by ranking the projects by preference. Among the possibilities were Irrlicht, Eucalyptus, Firefox, LibreOffice, and VLC. We kept the top 2 for consideration: Eucalyptus and Irrlicht,

Eucalyptus is cloud computing platform that bears many similarities to Amazon’s proprietary EC2 platform. Cloud computing is a huge industry buzz word lately that means a different thing depending on who you ask, so maybe as this semester goes on I will know to define it properly. The big advantage this project has is that our instuctor, Karl Wurst, knows 2 of the developers and they really want us to work on Eucalyptus and will offer us as much assistance as possible. There is an open source and a commercial version of Eucalyptus. Being the FOSS advocate that I am, it concerns me a bit that certain features are witheld and appear only in the commercial product. Maybe it’s because I have nightmares about Oracle and what they are doing to MySQL with open core. On the other hand, it’s great that a company is releasing a (mostly) fully featured product under the GPL 3 license. To work on this project we will be installing some flavor of Red Hat style OS (likely CentOS 6) onto 8 machines and making a cluster out of them. I have never made a cluster before so my inner system administrator is excited to get it up and running. Our goal will be to get a functional “cloud” environment set up running the latest and greatest Eucalyptus 3.0 code, which has not been released into the wild yet. The class will be writing installation documentation and bug reporting, but I’m keeping my fingers crossed that I will be getting some real commits into the code repository by the end of the semester.

The great Karl Wurst (can I have an A, please?) may let some students work on the Irrlicht project as well.  Irrlicht is a cross platform 3D library written in C++. The Irrlicht developers are interested in creating a test suite for benchmarking purposes, and they want my class to help write it. As far as resume material goes, I think Eucalyptus will be the more fruitful project. On the other hand, I do dabble in game development and have used a similar library before called OGRE, as well as used the OpenGL C API directly, so I am interested in this project as well. I don’t see myself actually being able to work on this, even if we are given the option as my time will surely be focused on Eucalyptus.

This class is making do all of the good things that I have been too lazy to learn how to use effectively on my own, like writing these blog posts and subscribing to listservs. Listservs are a great resource for project discussion that I never found convenient enough to actually use. Time to change that.