OpsWorks war deployment failure from S3 - aws-opsworks

Question

I have a war file,. myapp.war (it happens to be a grails app, but this is not material)
I upload this to an s3 bucket, say myapp in us-west-2
I set up an OpsWorks using the S3 repository type:
Repository Type: S3
Repository URL: https://myapp.s3-us-west-2.amazonaws.com/myapp.war
Access key ID: A key with read permission on the above bucket
Secret access key: the secret for this key
Deploy to an instance in Java layer (Tomcat 7)
All lights are green, deployments succeeded
But the app isn't actually deployed
Shelling in to the instance and looking in /usr/share/tomcat7/webapps I find a directory called 'myapp'. Inside this directory is a file called 'archive'. 'archive' appears to be a war file, but it is not named 'archive.war', and it is in a subdirectory of webapps, so tomcat isn't going to deploy it anyway.
Now, the OpsWorks docs say the archive should be a 'zip' file. But:
zipping up myapp.war into a zip archive 'myapp.war.zip' and changing the path to this file results in 'myapp' containing 'myapp.war'. No deployment, since tomcat isn't looking for war files in 'webapps/myapp'
Changing the name of 'myapp.war' to 'myapp.zip' and changing the repository path results in 'myapp' containing the single file 'archive' again.
So. Can anyone describe how to properly provide a war file to OpsWorks from S3?

Answer 1

It appears that the problem has to do with how the zip archive is made.
Jars, war, and the like created with the java 'jar' tool do not work. Zip archives created with a zip tool, and then renamed to have a '.war' extension do.
This is explained here: https://forums.aws.amazon.com/thread.jspa?messageID=559582&#559582
Quoting that post's answer:
Our current extract script doesn't correctly identify WAR files. If
you unpack the WAR file and use zip to pack it, it should work until
we update our script.
So the procedure that works is to:
Explode the war made by your development environment (In the case of grails, the war build cleans up the staging directory for the war, so you don't have an exploded war directory laying around to zip up yourself, you have to unzip it first.)
Zip the contents of the directory created by exploding the war using a zip tool (or, if your build tool leaves the exploded war directory there, then just zip it directly)
Optionally, rename the new zip archive to have a '.war' extension.
Resume the procedure from the original question, step 3 -- that is, upload the war to the s3 bucket and
Specify the S3 path to the war file as the repository in the OpsWorks setup.
EDIT:
After answering this, I discovered that Grails can produce an exploded war directory after all.
// BuildConfig.groovy
...
grails.project.war.exploded.dir = "path/to/exploded/war-directory"
grails.war.exploded=true
...
That directory can be zipped or jarred or whatever you want by your builder/deployer.

Answer 2

From this wiki page you see that a WAR file is just a special JAR file. And if you check out what a JAR is here then you see it is just zipped up compiled java code.
This SuperUser question also touches on the .WAR vs .zip business. Basically, a WAR is just a special ZIP. So when you upload a WAR, you are uploading a ZIP.
Make sure it's a WAR file in the S3 bucket.
Provide the entire link to the S3 WAR file. To get this, right-click the WAR file in S3 and select Properties and then copy the link.