Clean up and fix CPU issues

[lewfish]

This PR does various cleanup tasks, improve documentation, and fixes some issues re: the CPU resources. I tested this by creating the stack using cloudformation/template.yml following the README (the parameter values I used are posted in the #raster-vision Slack channel), and then using the instructions in azavea/raster-vision#668 (review) with the Vegas example.

After this is merged, I plan to rename raster-vision-aws to raster-vision-aws-terraform, and this repo to raster-vision-aws.

Closes #3
Closes #4
Closes #18
Closes azavea/raster-vision#512
Closes #19

[lewfish]

For review, I was hoping that @jeancochrane could look at the changes, and after it is approved, @jamesmcclain could test it by following the whole README.

[jamesmcclain]

For review, I was hoping that @jeancochrane could look at the changes, and after it is approved, @jamesmcclain could test it by following the whole README.

Yes

[jeancochrane]

Thanks for leading up this effort! The setup process is much clearer now, and I appreciate all of the troubleshooting details you added to the CloudFormation instructions.

Couple small comments on the docs and the template order, mostly nitpicks. I'll leave functionality testing to @jamesmcclain.

[lewfish]

Thanks; fixed according to suggestions.

[jamesmcclain]

I followed the instructions on a fresh virtual machine.

The instructions seem to be out of date: there is no create-ami target in the Makefile. Instead I had to build both the packer-image and create-image targets. I was looking at the right instructions but was on the wrong branch.

The instructions say that AWS_PROFILE environment variable must be set, but perhaps it should be emphasized that forgetting to set it will cause the create-ami build to hang indefinitely and have to be stopped from another terminal.

Also, the packer build fails with the latest Deep Learning AMI (Version 21):

That seems to be a well-known issue (see here and here) in various contexts. Using the Version 11 AMI from the screen shot works (for now), but that may not be true in the future and will not help people in other regions. I was not able to find AMIs for older versions from the UI.

The instructions gave me resources named RasterVisionCpuJobQueue, RasterVisionGpuJobQueue, RasterVisionHostedCpuJobDefinition, RasterVisionHostedGpuJobDefinition. These generic name are likely to collide if more than one user follows the instructions. I see some non-generic names (apparently) been created by the setup, but I did not see how to do that in the instructions (update: I found the option on my second run-through, but it does not seem to be clearly spelled-out in the instructions).

Finally, I have attempted to run the Vegas SpaceNet example, but the first CPU job is stuck in the RUNNABLE state. This could mean that that problem is unresolved, or it may be that I am simply using stale resources (see the previous paragraph).

[lewfish]

Also, the packer build fails with the latest Deep Learning AMI (Version 21):

I just had success running the packer job with Deep Learning AMI (Version 21). This makes me think it's an intermittent problem. I made a new issue for this #9 and made a note of it in the README.

I also added other warnings based on your feedback.

[jamesmcclain]

Also, the packer build fails with the latest Deep Learning AMI (Version 21):

I just had success running the packer job with Deep Learning AMI (Version 21). This makes me think it's an intermittent problem. I made a new issue for this #9 and made a note of it in the README.

I also added other warnings based on your feedback.

Sounds good