Finish Capistrano / get Yale ITS deployment set up

[orenyk]

I know that this isn't really a general application issue, but we need to get the ITS deployment procedure aligned with the changes implemented in #683. We can use this to finalize the implementation of Capistrano so that we hopefully have a well-documented example of how one might deploy an instance of Reservations.

---

2015-02-15

Here are some things that need to be done.

• Document USE_LDAP and ignore .env #1104
• have deploy.rb copy over the secrets.yml.example file
• review META: Heroku #275 branch to see what else generally needs to be done

[dgoerger]

@orenyk what's left to get Capistrano working? Does it work on e.g. laptops, if I were to git clone reservations and run a cap production deploy? Is there anything I can help with on the sysadmin side?

[orenyk]

Ummm... good question. It's set up to use localhost as the server but would require a hudson user at the moment. I haven't set up deploy.rb to copy over the secrets.yml file, so that would still be necessary, but other than that you'd have to set some environment variables and it should work. Maybe we should make the user configurable as an environment variable as well so that the whole Capistrano setup is as general as possible.

From ITS' side, we're basically going to need to copy all of the Jenkins parameters into a .env file and pass the deployment path to Capistrano as an environment variable as well. We should set up a meeting over the next month or so to hash this all out, but in the meantime look over #683 and the relevant wiki pages and let me know if anything jumps out at you.

[dgoerger]

I think the localhost/hudson and secrets.yml is something INF can replace with sed for example, much like we do now---and just dump the parameters into a .env perhaps instead of running it all as bash environment variables.

These are all the BMEC vars (available in /ror_dev/ruby_deploy/reservations01/install-PROD-BMEC_properties):

rails_environment=${deployment_env}
app_path=/path/to/install/directory
database=${bmec_db}
database_host=${which_server_hosts_bmec_db}
db_username=${bmec_db_username}
db_password=${bmec_db_password}
uri=/bmec

The Configuration doc looks pretty thorough. Does it work if one submits dummy values (e.g. localhost with Yale CAS and SMTP) for both the staging and production envs? It would be best to make sure that's working before INF gets involved---once it's confirmed working, we should definitely set up a meeting with Michael and get this going. 😀

[orenyk]

Ok, things are slowly moving. I've been trying to get this working on a DigitalOcean droplet running CentOS w/ Passenger and Apache (to get close to our setup in ITS) and it's... getting there 😕. Currently, I've got cap production deploy working until it tries to precompile assets when it fails because I haven't yet run the task that copies over database.yml. Hopefully I'll get it working soon and then we can hand it over to ITS to finish up.

One other thing to note: we've got pg in our production group in the Gemfile; this is so that we can deploy to Heroku. However, it also means that when you run bundle install you have to have postgresql and postgresql-devel (I think) installed on your app server so that it can install the pg gem, even if you're not using it. I'm not sure what the best way to handle this is, and whether or not ITS would be willing to just install those dependencies so that I don't need to figure out a workaround. We'll see...

[orenyk]

More to do:

• don't forget to have it working with localhost (I ran into issues with the ssh shell / rbenv not working, see here, rbenv seems to be problematic)
• take all the rbenv stuff out of deploy.rb and go back to rvm

[orenyk]

Ok, better but still stuck. I'm having trouble using the test commend from SSHKit, it's returning unknown command '['. Will investigate later.

[orenyk]

Ok, from memory ITS is using Passenger to deal with assets rather than through Rails, so I'm going to look into that now.

[orenyk]

Never mind, I had a relative subdirectory set in the configuration parameters (e.g. example.com/test) so the asset URL's were pointing to non-existent locations. For some reason the restart mechanism isn't working, nor is the whenever crontab generation task. The investigation continues...

[orenyk]

💥

[orenyk]

just testing party foul, but I think this is basically ready 😄

[orenyk]

Got it to work by tweaking the initializer to create errors for ActiveRecord::RecordNotFound exceptions and removing the rescue_from block in ApplicationController --> we have a successful deployment 😄. Now just to clean up and we'll go from there.

[orenyk]

Note that the current syntax would be GIT_TAG=v1.2.3 DEPLOY_DIR=/path/to/dir PARAM_FILE=/path/to/file cap production deploy

[orenyk]

PR opened, I'm just going to run through it's diff later to see if there's anything I'd like to clean up and then it should be ready for review / kicking up the ladder 😄.

[orenyk]

@dgoerger: any chance you could check this out and let me know if you think we're ready to set up a meeting w/ ITS? Thanks!

[orenyk]

... still waiting on ITS to start testing...

[orenyk]

I will be writing / updating our documentation shortly to include the use of Capistrano for deployment.

[orenyk]

Just realized that the whenever crontab update might not be running, might need an immediate patch to v5.3.0 😛. Just testing now, DigitalOcean is awesome!

[orenyk]

Whoops, I'm an idiot, I have it set to only call update_cron when it's deploying to production. Unfortunately that won't help ITS, so they might have to manually delete cron jobs on DEV, but I'm glad I set everything up properly 😛.

[orenyk]

On second thought, some of our cron tasks are not e-mail related, so we probably want them running on all environments. I'll open an issue for further investigation.