Merge lp://staging/~qzhang/lava-dispatcher/fix-867858 into lp://staging/lava-dispatcher

In my opinion, it's better to delete the directory ['/tmp/%s' % self.context.lava_result_dir] first before copy the bundles to the directory.

Then, the bundles in that directory will all be files from result_disk.
There will no garbage files in the target [/tmp] directory.

OK, let's make it simpler

I'm not sure it wouldn't be safer to just make sure the first thing we do before trying to deploy an image, is to reformat the partitions. There's a window of opportunity, where you could have a failure before the image deploy step gets to the point of formatting the partitions (for instance, if there's a failure during l-m-c). If anything is leftover from the previous run, then we could pick it up accidentally.

Add multiple emergency exits:)

Hi Spring, a couple of problems here still.

1. I don't think we need to format the partitions twice.
2. There's still a window of opportunity where we could fail earlier before the formatting takes place. For instance, if the network fails to come up. I would suggest moving the formatting into a _format_foo() method, call the logging bit to give info that we are formatting, and call it very very early in run().

I put the format just after system enters master image.
For Android test result will store on random directory after another bug fixed, I didn't handle Android part.

But still have one concern, does it need to umount the test partitions when deploying tarball for Ubuntu desktop image will mount it automatically if the formatting occurs before system enters Gnome.

+1