Using Opscode Chef to start up a node on AWS EC2 - A Simple Example

Chef from opscode.com is a suite of tools for managing computing infrastructure, from spinning up new nodes to installing Ruby gems to any custom operation you care to code up. It gets great reviews from those who manage numbers of unix (and other) systems and it has support for Amazon AWS and other cloud vendors baked right in.

I'm just getting started with it and am eager to go further but, like many projects, the documentation is far from clear. It has a lot of options and that typically means complexity. Perhaps because they are targeting folks who do systems administration for a living, their docs skim over some of the crucial basics.

My main interest in Chef is using it to spin up AWS EC2 nodes, configure them to my tastes and then spin them down as needed.

Here are the steps I took to get a node up and running:

0: I'm assuming you know how to fire up EC2 nodes from the management console or the command line on your desktop machine, that you have a keypair set up and that you are familiar with AMIs and the other terminology. If not, then get comfortable with EC2 before going any further.

1: I'm also assuming that you have a basic chef setup on your desktop and an account on Opscode's chef server (it's free for up to 5 nodes) and that you have written and tested a simple recipe on your desktop. If not, get comfortable with that before going further.

2: Add your AWS credentials to your knife.rb file (typically in ~/.chef)
The docs say to enter them in this format, which is different from the other configuration parameters that are generated from the opscode server... I don't know why... but this works:

# EC2 access keys
knife[:aws_access_key_id]     = "<your access key id>"
knife[:aws_secret_access_key] = "<your secret access key>"

3: Figure out your EC2 parameters for the node you want to spin up.

The AMI ID - I'm using a 32 bit Ubuntu AMI (ami-480df921)

The Instance Type - I'm testing with a Micro instance (t1.micro)

The Keypair name (mine is called 'craic-ec2-keypair') and the location of the SSH identity file that is linked with this (mine is in ~/.ssh/craic-ec2-keypair.pem).

The User that you (and chef) will ssh in to the node with (Ubuntu 10.04 wants me to use 'ubuntu' instead of 'root')

The Security Group for the server (I just use 'default' which is what I set up when I first started with EC2)

The Chef Recipe(s) and/or Role(s) that you want to run on the new EC2 server (I'm just going to run a simple one that I created called 'craic_test' that writes a file in /tmp) - Make sure this works on a non-EC2 client first!

4: Let's do it...
The chef command that you use is 'knife' with some EC2 specific options. You can find more information HERE in their FAQs, which is MUCH more complete than their Wiki documentation on knife...

The command will be 'knife ec2 server create' followed by a bunch of options. That's pretty self explanatory - but then we get into the options... Here is the full command I used split across multiple lines, which I shall explain below:

$ knife ec2 server create "recipe[craic_test]" \
      -i ami-480df921 \
      -f t1.micro \
      -x ubuntu \
      -S craic-ec2-keypair \
      -I ~/.ssh/craic-ec2-keypair.pem \
      -G default

The first line is the command itself followed by the recipes and/roles in the standard chef format ("recipe[craic_test]").

The EC2 AMI is specified with the -i option, then the Instance Type is specified with the -f option (aka --flavor - I don't know why they chose that...).

Next comes the ssh user with the -x option, the SSH Keypair name with -S (upper case S) and the location of the matching identity file with the -I option (upper case I) - then finally the security group with the -G option (upper case G).

I am really hoping that there is some way to store some of these options in knife.rb or someplace convenient as most of these will not change for my uses.

All being well, what happens next is that the node gets spun up on EC2, chef installs the prerequisites it needs to run on the node and downloads the chef client over there. Chef will create the new node record on the chef server, download your recipes and then execute them.

What you will see is a TON of output as everything runs. I won't subject you to all of that here, but the first few line will look like this:

Instance ID: i-08da8865
Flavor: t1.micro
Image: ami-480df921
Availability Zone: us-east-1b
Security Groups: default
SSH Key: craic-ec2-keypair

Waiting for server............
Public DNS Name: ec2-184-72-190-66.compute-1.amazonaws.com
Public IP Address: 184.72.190.66
Private DNS Name: domU-12-31-39-0F-29-E8.compute-1.internal
Private IP Address: 10.193.42.22

Waiting for sshddone
INFO: Bootstrapping Chef on
0% [Working]90-66.compute-1.amazonaws.com
[...]

This shows the IP info, etc for your new instance and is followed by pages of messages reporting the downloading and installation of all the prerequisites for Chef. At the end you should see your recipes being run and finally a summary once again of the instance information:

Instance ID: i-08da8865
Flavor: t1.micro
Image: ami-480df921
Availability Zone: us-east-1b
Security Groups: default
SSH Key: craic-ec2-keypair
Public DNS Name: ec2-184-72-190-66.compute-1.amazonaws.com
Public IP Address: 184.72.190.66
Private DNS Name: domU-12-31-39-0F-29-E8.compute-1.internal
Private IP Address: 10.193.42.22
Run List: recipe[craic_test]

5: Now you can ssh into your node and you should see that your recipes have run and done whatever you asked of them.

One good thing about all the prerequisites getting installed, for an Ubuntu system at least, is that you now have a node with a lot of development libraries and tools already in place (especially if you work in Ruby).

That's it for the basic installation.

Remember to terminate your node when you are finished as you are being charged for it AND remove any test nodes from your Opscode server account as they will count against your account limit.

There is obviously a lot more that you can, and need to, do to set up your nodes but this walk through gets you to a functioning node.

Happy cooking

System-wide rvm install on Ubuntu

Here are the steps I used to install RVM system wide on a Ubuntu 10.04.1 LTS node on Amazon EC2, using one of the Canonical AMIs. These set up the primary user as 'ubuntu', requiring you to sudo for system commands. I could install rvm under that user but installing System-Wide is recommended.

This HOWTO is based on the RVM page http://rvm.beginrescueend.com/deployment/system-wide/. But I find the RVM site documentation makes a few assumptions with regard to commands.

1: Install git if you don't already have it

$ sudo apt-get install git-core

2: Download RVM

$ sudo bash < <( curl -L http://bit.ly/rvm-install-system-wide )

This creates /usr/local/rvm and the group 'rvm'

3: Add each user to group rvm

$ sudo usermod -aG rvm ubuntu

4: Add startup lines to each user's .profile

[[ -s '/usr/local/lib/rvm' ]] && source '/usr/local/lib/rvm'

5: Add this line to /etc/rvmrc (may not be necessary)

$ echo "export rvm_path=/usr/local/rvm" > /etc/rvmrc

6: Use rvm as a user - best to logout/login first to make sure you get all the paths etc setup

$ rvm list
$ rvm package install zlib   ( I needed this to get install to work )
$ rvm install 1.9.2 --with-zlib-dir=/usr/local/rvm/usr

Note that you use rvm as a regular user, not via sudo, even though this is a system wide install.


 

Extracting Text from PDF documents on Mac OS X

There are various ways to extract the text from a PDF document. In Mac OS X there are two effective ways that cost nothing.

1: Open the PDF in Preview or Adobe Acrobat Reader, select all the text, copy and paste into your editor of choice.

This is a quick and easy solution for one or a few documents. But for a large number of documents it would quickly become tedious.

Neither application allows you to Save As Text from the menu, which is unfortunate.

I also see a significant difference between the pasted text when the original PDF has columns of text. Compare the same original text copied from Acrobat Reader:

MOLECULAR FORMULA C6492H10060N1724O2028S42
MOLECULAR WEIGHT 146.0 kDa
TRADEMARK None as yet
MANUFACTURER MedImmune
CODE DESIGNATION MEDI-563
CAS REGISTRY NUMBER 1044511-01-4

and Mac OS X Preview:

MOLECULAR FORMULA MOLECULAR WEIGHT TRADEMARK MANUFACTURER CODE DESIGNATION CAS REGISTRY NUMBER
C6492H10060N1724O2028S42 146.0 kDa None as yet MedImmune
MEDI-563 1044511-01-4

The first version is the right one. So copying from Acrobat Reader is the better solution.

2. To handle many documents you want a script of some sort that just extracts the text. You can do this with the Mac OS X Automator application. I'd not used this before but it gives you access to a whole load of functions/services. Well worth checking out. You'll find Automator in your applications folder. Open it up and...

- Select Application in the template window that appears on opening.
- In the Automator window you have two vertical panels on the left and a grey workspace on the right. Click the 'Actions' button in the top left.
- In the left hand column, under Library, click 'PDFs' to bring up a list of PDF related actions in the second column.
- Select, drag and drop 'Extract PDF Text' onto the workspace. A dialog will appear.
- Select the appropriate options - plain text or RTF, folder in which to output the text, etc.
- Save As... and give your new app a suitable name
- Quit Automator
- Drag and drop a PDF onto the icon for the app and it will extract the text and output that in a file in the directory you configured in the app

Simple! Now you can just drag and drop PDFs and they will be converted automatically.

Automator can do a whole lot more than this, especially if you have a series of steps that must applied to files (e.g. image processing). If you are into scripting, take a look at the way this all works. Right click on the app icon and 'Show Package Contents'.

You can also learn about this via this article by Mathias Bynens on a simple way to convert UNIX scripts into apps ... it's the same mechanism.

So now I've got an automated way to extract text... BUT... the script uses Preview for the extraction, and that still gives me the incorrect ordering of text that I showed in the above example... dang it... not a deal breaker but frustrating all the same


 

Javascript Date.parse browser compatibility issue

Just got burned with a Javascript incompatibility between Firefox and Safari on the Mac...

Date.parse() takes a string representation of a Date and return the number of milliseconds since the epoch. It has always been able to take dates in IETF format, such as "Jan 1, 2010" but as of Javascript 1.8.5 it can handle ISO8601 format as well, such as "2010-01-01".

I use ISO8601 in all my applications and in Firefox 3.6.12 (on the Mac) this works fine:

> Date.parse("Jan 1, 2010 GMT");
1262304000000
> Date.parse("2010-01-01");
1262304000000

Note that the ISO8601 form assumes the GMT timezone, but the IETF form assumes local timezone unless you specify GMT.

The problem for me arose in Safari (5.0.2) on the Mac:

> Date.parse("Jan 1, 2010 GMT");
1262304000000
Date.parse("2010-01-01");
NaN

Hmm...

And just to muddy the waters further, here is the output on Google Chrome 7.0.517.44:

> Date.parse("Jan 1, 2010 GMT");
1262304000000
Date.parse("2010-01-01");
1262332800000

The ISO8601 date is handled OK but is returned in the local timezone.

I need a fail safe way to parse ISO8601 dates - what to do...? Here's what I came up with:

var date_str = '2010-01-01';
var iso8601_regex = /(\d{4})[\/-](\d{2})[\/-](\d{2})/;
var match = iso8601_regex.exec(date_str);
var date = new Date(match[1], match[2] - 1, match[3]);
var milliseconds = Date.parse(date);    // -> 1262332800000 (local time)
milliseconds = Date.UTC(match[1], match[2] - 1, match[3], 0, 0, 0);  // -> 1262304000000 (UTC/GMT)

Note the '- 1' with the month (match[2]) - the first month is 0 - go figure.
This produces the same results in Mac Firefox, Safari and Chrome - consistent behaviour - yes, really!

What a palaver...

 

Rails has_many :through associations and edit/update actions

Consider a basic has_many :through association

class Drug < ActiveRecord::Base
  has_many :indications
  has_many :diseases, :through => :indications, :uniq => true

Here a drug can be used to treat multiple diseases and any disease can be treated with multiple drugs (that side of the association is not shown). The 'indications' is a basic linking table with drug_id and disease_id.

In the drug#new and drug#edit forms you might use a select menu that allows you to select multiple diseases. I use the 'simple_form' gem for my forms and the 'association' method makes this trivial.

<%= f.association :diseases,
           :collection => Disease.all(:order => 'name') %>

In order for this to work you need to add an attr_accessible called :disease_ids to your model. With that, simple form should handle all details needed to create and update the association.

But there is a problem with the edit/update actions when you want to deselect ALL diseases. If you do this in the form then no disease_ids parameter will get passed to your controller and so this column will not get updated. It is a classic issue with HTML form updates and applies to checkboxes as well.

The solution is to add a line to the updater action in your controller that sets the disease_ids parameter to an empty array if it does not exist:

  def update
    params[:drug][:disease_ids] ||= []
    @drug = Drug.find(params[:id])
    [...]

This works fine - adds a bit of clutter to the controller but there you go...

However, this will break a basic functional test for the update action and you will get an error similar to this:

test_update_valid(DrugsControllerTest):
NoMethodError: You have a nil object when you didn't expect it!
You might have expected an instance of Array.
The error occurred while evaluating nil.[]

The problem stems from the basic stub for your object not passing a params[:drug] hash. Now, I'm not an expert on stubs/mocking so there may be a much cleaner way of fixing this, but I fix this by explicitly creating the needed parameters and passing them in the 'put' method.

Here is an example of a basic update test

  def test_update_valid
    Drug.any_instance.stubs(:valid?).returns(true)
    put :update, :id => Drug.first
    assert_redirected_to drug_url(assigns(:drug))
  end

and here is the modified one that will work

  def test_update_valid
    Drug.any_instance.stubs(:valid?).returns(true)
    put :update, { :id => Drug.first, :drug => { :disease_ids => [] } }
    assert_redirected_to drug_url(assigns(:drug))
  end

Of course I should be building out the tests to provide truly useful tests, but if you can't get beyond this step, the others don't really matter.

Hope this helps....

 

Disabling spell check in HTML forms

I work with DNA and protein sequences and I often have HTML forms with a textarea for entering sequence. Unfortunately my browser sees that text (e.g. 'agctagagctcgatagc') and decides that this is misspelled and underlines all the sequence text with a red dotted line... ugly...

In HTML5 you can now disable spell checking on textarea and text inputs using the option 'spellcheck' = 'false' - EASY!

Note that this is a HTML attribute, NOT CSS - so you have to set it in the form itself.

Browser support for the feature may vary. It works on Firefox and Safari on the Mac for sure.

A related attribute is 'contenteditable' that allows you to control whether specific parts of a textarea's content can be modified - like 'readonly' but with much more control.

 

Rails, Factory_girl and Bioinformatics - a gotcha

I've been using factory girl as a replacement for Fixtures in a new Rails application. It has been working well but I just stumbled across a BIG gotcha for my application.

I work on bioinformatics applications with DNA and Protein sequences and having a model with a column called 'sequence' is a natural choice.

The problem is that Factory Girl allows you to create objects in which a specific field is given a sequential id. For example 'user_1', 'user_2', etc. You set that up in your Factory definition like this:

Factory.define :mymodel do |f|
  f.sequence(:name) {|n| "name_#{n}" }
end

When you have a model with a field called sequence you would define that like this:

Factory.define :mymodel do |f|
  f.sequence(:name) {|n| "name_#{n}" }
  f.sequence 'acgtacgtacgt'
end

You see the problem... Running a test with this in it brings up an error message like this:

  1) Error:
test: A Sequence instance should be valid. (SequenceTest):
NoMethodError: undefined method `call' for nil:NilClass
    /Users/jones/.rvm/gems/ruby-1.9.2-p0/gems/activesupport-3.0.1/lib/active_support/whiny_nil.rb:48:in `method_missing'

The same problem will arise if you have a column called 'association'.

This is a bad design choice to mix up field names and methods. A better choice would have been to use a 'verb' such as f.generate_sequence(:name), which is much less likely to be used as a column name.

I can't see a way around this other than changing the name of the column in my model, which I am very reluctant to do.
Machinist is an alternative to Factory Girl, so that might be a solution - or hacking the factory_girl gem to change the name of the methods...

UPDATE: Factory Girl will allow alternate syntaxes - see the "Alternate Syntaxes" section. Not sure if its the best path for me but might get me over the immediate hurdle.

UPDATE 2: Just tried Machinist and that solves the first problem... however it fails if you have a column called 'alias' in a blueprint - the trick here is to precede the column name with 'self', i.e. 'self/alias' works.