Running integration tests on ephemeral server instance using Heroku CI - node.js

Question

I'm trying to take advantage of this new feature of Heroku to test a parse-server/nodejs application that we have on Heroku, using mocha.
I was expecting Heroku to launch an ephemeral instance of my app along with the tests so that they could be run against it, but it doesn't seem like that's happening. Only the tests get launched.
Now, I found at least one snippet about configuring the Dyno formation to use dynos other than performance-m for the test, so I'm trying to declare my other dynos there as well:
"environments": {
"test": {
"scripts": {
"test-setup": "echo done",
"test": "npm run test"
},
"addons": [
{
"plan": "rediscloud:30",
"as": "REDISCLOUD_URL"
}
],
"formation": {
"test": {
"quantity": 1,
"size": "standard-1x"
},
"worker": {
"quantity": 1,
"size": "standard-1x"
},
"web": {
"quantity": 1,
"size": "standard-1x"
}
}
}
}
in my app.json, but it seems to be getting totally ignored.
I know my mocha script could import the relevant part of the web server and test against it, and that's what I've seen in the non-heroku-related examples, but our app consists of a worker too, and I'd like to profile the interaction of both and test the job lengths against our expectations of performance, rather than individual components, hence "integration tests". Is this a legit use for Heroku tests or I'm doing something wrong or have wrong expectations? I'm more concerned about this than getting it to work, because I'm quite certain I could get it to work in a certain number of ways (mocha spawning the server processes, npm concurrently package, etc), but if I can avoid hacks, all the better.
Locally, I was able to get both imported in the script, but the performance is degraded since it's now 2 processes + the tests running in a single memory process, with nodejs's memory cap limitations and a single event loop instead of 3. While writing this I'm thinking I could probably use throng and spawn different functions depending on the process ID. I'll try this if I don't get any better solutions.
Edit: I already managed to make it run by spawning the server/worker as separate processes in a before step in mocha, calculating the proper ram amounts to allow to each using the env vars. I'm still interested in knowing if there's a better solution.