Jay Harris is Cpt. LoadTest

a .net developers blog on improving user experience of humans and coders
Home | About | Speaking | Contact | Archives | RSS
 

Azure Websites are a fantastic method of hosting your own web site. At Arana Software, we use them often, particularly as test environments for our client projects. We can quickly spin up a free site that is constantly up-to-date with the latest code using continuous deployment from the project’s Git repository. Clients are able to see progress on our development efforts without us having to worry about synchronizing codebases or managing infrastructure. Since Windows Azure is a Microsoft offering, it is a natural for handling .NET projects, but JavaScript-based nodejs is also a natural fit and a first-class citizen on the Azure ecosystem.

Incorporating Grunt

Grunt is a JavaScript-based task runner for your development projects to help you with those repetitive, menial, and mundane tasks that are necessary for production readiness. This could be running unit tests, compiling code, image processing, bundling and minification, and more. For our production deployments, we commonly use Grunt to compile LESS into CSS, CoffeeScript into JavaScript, and Jade into HTML, taking the code we write and preparing it for browser consumption. We also use Grunt to optimize these various files for speed through bundling and minification. The output of this work is done at deployment rather than development, with only the source code committed into Git and never its optimized output.

Git Deploy and Kudu

Continuous deployment will automatically update your site with the latest source code whenever modifications are made to the source repository. This will also work with Mercurial. There is plenty of existing documentation on setting up Git Deploy in Azure, so consider that a prerequisite for this article. However, Git Deploy, alone, will only take the files as they are in source, and directly deploy them to the site. If you need to run additional tasks, such as compiling your .NET source or running Grunt, that is where Kudu comes in.

Kudu is the engine that drives Git deployments in Windows Azure. Untouched, it will simply synchronize files from Git to your /wwwroot, but it can be easily reconfigured to execute a deployment command, such as a Windows Command file, a Shell Script, or a nodejs script. This is enabled through a standardized file named ".deployment". For Grunt deployment, we are going to execute a Shell Script that will perform npm, Bower, and Grunt commands in an effort to make our code production-ready. For other options on .deployment, check out the Kudu project wiki.

Kudu is also available locally for testing, and to help build out your deployment scripts. The engine is available as a part of the cross-platform Windows Azure Command Line Tools, available through npm.

Installing the Azure CLI

npm install azure-cli –-global

We can also use the Azure CLI to generate default Kudu scripts for our nodejs project. Though we will need to make a few modifications to make the scripts work with Grunt, it will give us a good start.

azure site deploymentscript –-node

This command will generate both our <code>.deployment</code> and the default <code>deploy.sh</code>.

Our .deployment file

[config]
command = bash ./deploy.sh

Customizing deploy.sh for Grunt Deployment

From .deployment, Kudu will automatically execute our deploy.sh script. Kudu’s default deploy.sh for a nodejs project will establish the environment for node and npm as well as some supporting environment variables. It will also include a "# Deployment" section containing all of the deployment steps. By default, this will copy your repository contents to your /wwwroot, and then execute npm install --production against wwwroot, as if installing the application's operating dependencies. However, under Grunt, we want to execute tasks prior to /wwwroot deployment, such as executing our Grunt tasks to compile LESS into CSS and CoffeeScript into JavaScript. By replacing the entire Deployment section with the code below, we instruct Kudu to perform the following tasks:

  1. Get the latest changes from Git (or Hg). This is done automatically before running deploy.sh.
  2. Run npm install, installing all dependencies, including those necessary for development.
  3. Optionally run bower install, if bower.json exists. This will update our client-side JavaScript libraries.
  4. Optionally run grunt, if Gruntfile.js exists. Below, I have grunt configured to run the Clean, Common, and Dist tasks, which are LinemanJS's default tasks for constructing a production-ready build. You can update the script to run whichever tasks you need, or modify your Gruntfile to set these as the default tasks.
  5. Finally, sync the contents of the prepared /dist directory to /wwwroot. It is important to note that this is a KuduSync (similar to RSync), and not just a copy. We only need to update the files that changed, which includes removing any obsolete files.

Our deploy.sh file's Deployment Section

# Deployment
# ----------

echo Handling node.js grunt deployment.

# 1. Select node version
selectNodeVersion

# 2. Install npm packages
if [ -e "$DEPLOYMENT_SOURCE/package.json" ]; then
  eval $NPM_CMD install
  exitWithMessageOnError "npm failed"
fi

# 3. Install bower packages
if [ -e "$DEPLOYMENT_SOURCE/bower.json" ]; then
  eval $NPM_CMD install bower
  exitWithMessageOnError "installing bower failed"
  ./node_modules/.bin/bower install
  exitWithMessageOnError "bower failed"
fi

# 4. Run grunt
if [ -e "$DEPLOYMENT_SOURCE/Gruntfile.js" ]; then
  eval $NPM_CMD install grunt-cli
  exitWithMessageOnError "installing grunt failed"
  ./node_modules/.bin/grunt --no-color clean common dist
  exitWithMessageOnError "grunt failed"
fi

# 5. KuduSync to Target
"$KUDU_SYNC_CMD" -v 500 -f "$DEPLOYMENT_SOURCE/dist" -t "$DEPLOYMENT_TARGET" -n "$NEXT_MANIFEST_PATH" -p "$PREVIOUS_MANIFEST_PATH" -i ".git;.hg;.deployment;deploy.sh"
exitWithMessageOnError "Kudu Sync to Target failed"

These commands will execute bower and Grunt from local npm installations, rather than the global space, as Windows Azure does not allow easy access to global installations. Because bower and Grunt are manually installed based on the existence of bower.json or Gruntfile.js, they also are not required to be referenced in your own package.json. Finally, be sure to leave the –no-color flag enabled for Grunt execution, as the Azure Deployment Logs will stumble when processing the ANSI color codes that are common on Grunt output.

Assuming that Git Deployment has already been configured, committing these files in to Git will complete the process. Because the latest changes from Git are pulled before executing the deployment steps, these two new files (.deployment and deploy.sh) will be available when Kudu is ready for them.

Troubleshooting

Long Directory Paths and the 260-Character Path Limit

Though Azure does a fantastic job of hosting nodejs projects, at the end of the day Azure is still hosted on the Windows platform, and brings with it Windows limitations. One of the issues that you will quickly run into under node is the 260-Character Path Limitation. Under nodejs, the dependency tree for a node modules can get rather deep. And because each dependency module loads up its own dependency modules under its child folder structure, the folder structure can get rather deep, too. For example, Lineman requires Testem, which requires Winston, which requires Request; in the end, the directory tree can lead to ~/node_modules/lineman/node_modules/testem/node_modules/winston/node_modules/request/node_modules/form-data/node_modules/combined-stream/node_modules/delayed-stream, which combined with the root path structure, can far exceed the 260 limit.

The Workaround

To reduce this nesting, make some of these dependencies into first-level dependencies. With the nodejs dependency model, if a module has already been brought in at a higher level, it is not repeated in the chain. Thus, if Request is made as a direct dependency and listed in your project's project.json, it will no longer be nested under Winston, splitting this single dependency branch in two:

  1. ~/node_modules/lineman/node_modules/testem/node_modules/winston
  2. ~/node_modules/request/node_modules/form-data/node_modules/combined-stream/node_modules/delayed-stream

This is not ideal, but it will solve is a workaround for the Windows file structure limitations. The element that you must be careful of is with dependency versioning, as you will need to make sure your package.json references the appropriate version of your pseudo-dependency; in this case, make sure your package.json references the same version of Request as is referenced by Winston.

To help find those deep dependencies, use npm list. It will show you the full graph on the command line, supplying a handy visual indicator.

__dirname vs Process.cwd()

In the node ecosystem, Process.cwd() is the current working directory for the node process. There is also a common variable named __dirname that is created by node; its value is the directory that contained your node script. If you executed node against a script in the current working directory, then these values should be the same. Except when they aren't, like in Windows Azure.

In Windows Azure, everything is executed on the system drive, C:. Node and npm live here, and it appears as though your deployment space does as well. However, this deployment space is really a mapped directory, coming in from a network share where your files are persisted. In Azure's node ecosystem, this means that your Process.cwd() is the C-rooted path, while __dirname is the \\10.whatever-rooted UNC path to your persisted files. Some Grunt-based tools and plugins (including Lineman) will fail because that it will reference __dirname files while Grunt's core is attempting to run tasks with the scope of Process.cwd(); Grunt recognizes that it's trying to take action on \\10.whatever-rooted files in a C-rooted scope, and fails because the files are not in a child directory.

The Workaround

If you are encountering this issue, reconfigure Grunt to work in the \\10.whatever-rooted scope. You can do this by setting it's base path to __dirname, overriding the default Process.cwd(). Within your Gruntfile.js, set the base path immediately within your module export:

module.exports = function (grunt) {
  grunt.file.setBase(__dirname);
  // Code omitted
}

Unable to find environment variable LINEMAN_MAIN

If like me, you are using Lineman to build your applications, you will encounter this issue. Lineman manages Grunt and its configuration, so it prefers that all Grunt tasks are executed via the Lineman CLI rather than directly executed via the Grunt CLI. Lineman's Gruntfile.js includes a reference to an environment variable LINEMAN_MAIN, set by the Lineman CLI, so that Grunt will run under the context of the proper Lineman installation, which is what causes the failure if Grunt is executed directly.

The Fix (Because this isn't a hack)

Your development cycle has been configured to use lineman, so your deployment cycle should use it, too! Update your deploy.sh Grunt execution to run Lineman instead of Grunt. Also, since Lineman is referenced in your package.json, we don't need to install it; it is already there.

Option 1: deploy.sh

# 4. Run grunt
if [ -e "$DEPLOYMENT_SOURCE/Gruntfile.js" ]; then
  ./node_modules/.bin/lineman --no-color grunt clean common dist
  exitWithMessageOnError "lineman failed"
fi

Recommendation: Since Lineman is wrapping Grunt for all of its tasks, consider simplifying lineman grunt clean common dist into lineman clean build. You will still need the --no-color flag, so that Grunt will not use ANSI color codes.

The Alternate Workaround

If you don't want to change your deploy.sh—perhaps because you want to maintain the generic file to handle all things Grunt—then as an alternative you can update your Gruntfile.js to specify a default value for the missing LINEMAN_MAIN environment variable. This environment variable is just a string value passed in to node's require function, so that the right Lineman module can be loaded. Since Lineman is already included in your package.json, it will already be available in the local /node_modules folder because of the earlier npm install (deploy.sh, Step #2), and we can pass 'lineman' into require( ) to have Grunt load the local Lineman installation. Lineman will then supply its configuration into Grunt, and the system will proceed as if you executed Lineman directly.

Option 2: Gruntfile.js

module.exports = function(grunt) {
  grunt.file.setBase(__dirname);
  if (process.env['LINEMAN_MAIN'] === null || process.env['LINEMAN_MAIN'] === undefined) {
    process.env['LINEMAN_MAIN'] = 'lineman';
  }
  require(process.env['LINEMAN_MAIN']).config.grunt.run(grunt);
};

Credits

Thank you to @davidebbo, @guayan, @amitapl, and @dburton for helping troubleshoot Kudu and Grunt Deploy, making this all possible.

Changelog

2013-12-03: Updated LINEMAN_MAIN Troubleshooting to improve resolution. Rather than editing deploy.sh to set the environment variable, edit the file to execute Lineman. This is the proper (and more elegant) solution. [Credit: @searls]

Technorati Tags: ,,,,
Tuesday, 03 December 2013 00:34:25 (Eastern Standard Time, UTC-05:00)  #    Comments [4] - Trackback