Use git caching to speed build times

It appears we are using S3 caching for software artifacts but not necessarily git caching to speed builds:

http://www.rubydoc.info/github/opscode/omnibus/Omnibus/Config#use_s3_caching-instance_method

For our full releases, we may still opt to do full rebuilds without git caching. But for quick branch deploys, we may want to use git caching.

mentioned in merge request !845 (merged)

@stanhu I did some work on this the other day and I have been testing how it is working. You can see the WIP MR here !845 (merged)

mentioned in commit fd12dbc2

mentioned in commit dc40efd6

mentioned in commit faeee516

As I expected, this only has limited impact. It slashes time on machine that is not newly provisioned, if it is then it starts building from the start. This is mostly due to preparation definition, which we could remove even. Testing in a separate branch.

Well no, if I extract the preparation definition, the next one is going to dirty the cache (package-scripts) because it needs to create some templates.

Bottom line here is, whatever we do on a new machine will 'dirty' the cache and the build will start from the beginning. We can try to fight this by increasing the reserved number of machines but that will add significant costs because of the sizes of the machines used.

mentioned in issue #1380 (closed)

mentioned in commit df54aa58

mentioned in commit e83bfe10

mentioned in commit 879d9354

mentioned in commit b481a92a

Closing for #1356 (closed)

Status changed to closed