Feedback on AWS deployment instructions

[rosenhouse]

Hey,

We've attempted to deploy Diego to AWS from scratch following these instructions. Here's a short (we hope) list of issues that we ran into:

1. boosh is not godep gettable. It references github.com/fraenkel/candiedyaml which does not exist. We had to manually change the Godeps.json and fix package imports in order to get it to build. This is an open PR in boosh.

2. In the "AWS keypair for the BOSH director" step, it says to name the keypair "bosh". We'd really like to be able to give this keypair the same name as our deployment -- so that we can distinguish it from other keypairs for other systems. But we're afraid to do this -- it seems like it might break something later on.

3. The same section should probably end with the instruction to run

   chmod 600 keypair/id_rsa_bosh
   

4. The Route53 Hosted Zone instructions are missing required steps to create the NS record in the parent domain that points into the new hosted zone.

@sykesm && @rosenhouse

[cf-gitbot]

We have created an issue in Pivotal Tracker to manage this. You can view the current status of your issue at: https://www.pivotaltracker.com/story/show/116088089.

[emalm]

Thanks, @rosenhouse and @sykesm! I've +1'd the boosh PR, and I'll pull the associated story in for the Diego team to fix the other issues.

For the last point, I don't think there's a universal set of instructions for all DNS providers (I had to do something else at my own registrar when I did acceptance on these instructions), but we can still describe the process of wiring the Hosted Zone delegation set in as the NS records for a subdomain in another Route 53 registration.

Best,
Eric

[sykesm]

One more small thing we hit while generating the Diego deployment manifest:

• "Instance-Count-Overrides" and "Release-Versions" are both marked as "(Optional)" so we assumed we could skip those steps. But then the generate-deployment-manifest script choked because the files were missing.

Maybe ensure people copy empty stubs to the deployments directory and optionally edit them?

Thanks.
@sykesm & @rosenhouse

[sykesm]

One more:

examples/aws/stubs/cf/diego.yml overrides the smoke_tests and acceptance_tests instance counts to 0. We're not sure why. It complicated our efforts to validate the deployment, and doesn't appear to save us any resources in the steady-state, given that BOSH creates errand VMs on demand.

[sykesm]

On that note: we found that the generated manifests were missing properties required to run the test errands.

[sykesm]

We are deploying to us-west-2 and found one more place where the AWS region needs to be overridden:

in stubs/cf/properties.yml we had to add

meta:
   fog_config:
      region: us-west-2

[luan]

Hey @sykesm and @rosenhouse: I pushed commits to develop addressing most of the issues you pointed out. I purposefully skipped the one about Route53 "parent" domains, I see that as a specific detail that can vary too much depending on how you're set up. For example we have a parent domain that is in a separate account, @ematpl seems to have the parent domain on a totally separate registrar while my personal account is parked within the same AWS Route53 setup. So based on that we decided to leave that part a little more open ended.

Let me know if the fixes pushed to develop address your concerns effectively!

Thanks,
@luan

[rosenhouse]

@luan not sure if you see it, but I left a comment here: db09bb1

Otherwise this looks good, thanks for all these fixes!

[luan]

Yeah, I saw that. Working on making that configurable now, thanks!

On Wed, Mar 23, 2016, 08:37 Gabe Rosenhouse notifications@github.com
wrote:

@luan https://github.com/luan not sure if you see it, but I left a
comment here: db09bb1
db09bb1

Otherwise this looks good, thanks for all these fixes!

—
You are receiving this because you were mentioned.

Reply to this email directly or view it on GitHub
#147 (comment)