Ask HN: Heroku at scale? | Hacker News Reader

Ask HN: Heroku at scale?

6 pointsian08 comments

Ive been a huge fan of Heroku since I first used it a few years back. Its amazing how quickly you can get up and running, deploy, add SSL, monitoring etc. The polar opposite of AWS. It has worked for me for a variety of applications, my only issues being their region support as it doesn't match up with where our customers are, adding latency to the app.

However more than one person has mentioned to me that its not a suitable platform to scale and that eventually we will run into issues. I do notice that they have quite a few outages in the past few months, but I have no benchmark really to compare to running say direct on AWS.

Does anyone here have any experience scaling relatively large apps (say dozens of dynos) on Heroku? Or perhaps mission critical apps? Did you run into any issues? Or share these concerns?

As background we are running a B2B business with a few different websites and mobile backends. Various languages/frameworks. There is a transactional component. Relatively low RPS.

I have experience running an application with 9M users (around 2M active) on Heroku. We had around 1,5k requests/second average, peaking at over 2k reqs/sec. No major issues, although we were starting to hit the per-app dyno limit (we could not go over 20 dynos) but this problem could be solved by creating a second app.

Overall no issues and I would definitely recommend them.

That's amazing! What does that cost per month? (if I may ask)

Only 20 dynos? I assumed Heroku could scale to hundreds of dynos per app deployment, as they assign a random local port to every dyno.

You could create a second app, but in certain cases, that violates their terms of service, where you're not allowed to use many deployments to act as a single webapp.

> What does that cost per month? (if I may ask)

I don't actually remember but I did the test on my own Heroku apps now and with Performance-M dynos (the ones we were using) we were looking at $5000+/month.

To be fair the most expensive things were the add-ons; we were spending over $10000/month for those.

Cheers - thats very comforting.

I think the problem with heroku is not scaling but the cost. They use AWS behind the scenes so you are paying a premium over AWS. Eventually that will become expensive. But if you are saving money on labor it's worth it.

>> But if you are saving money on labor it's worth it.

Yeah we definitely are. Labor and just plain hassle. It's great to be able to deploy in seconds, rollback easily. Of course its possible to set that up on AWS too - just in practice its always far more complicated than it seems.

The billing is very easy to understand, which is nice. Ive seen AWS billing spiral out of control, yet to experience that on Heroku (touch wood).

As far as scaling Heroku, it's usually the cost that will limit you, rather than the number of Dynos. Some alternatives, arguably cheaper:

- Dokku: a Docker-based Heroku clone, self-hosted (cheap!), compatible with Heroku build-packs, easy migration.

- Firebase: Google's answer to an all in one deployment solution, scales really well, fairly cheap too.

My view is; Heroku is a great option it's probably the fastest development cycle you could possible get. Instead of preempting scaling problems, stick with what you have, then migrate slowly to something bigger IF it becomes an issue.

> Instead of preempting scaling problems, stick with what you have, then migrate slowly to something bigger IF it becomes an issue.

Cheers - Yep that was our thinking too.

Reply on news.ycombinator.com