Hacker News Reader

Comment by capableweb | Hacker News Reader

Heroku is a general apps platform. This seems to be tailored specifically to Django/Python (with the option of using your own Dockerfile so could deploy anything). One could easily imagine that the performance and stability should be a lot higher than with Heroku. If you're a Python-only shop, it might make more sense.

View on HN

> One could easily imagine that the performance and stability should be a lot higher than with Heroku.

No, I can’t imagine that. Why would performance be any different? I imagine most of these kinds of solutions use Docker behind the hood anyway, so they’re likely to be identical.

Because Heroku is phenomenally expensive.

As i said in another comment Heroku is unbelievably expensive.

What they offer for $250 per process can be done at $60/mo per whole app.

When I helped them move all background workers and all non-production apps to Digital Ocean they ended up saving $7k/mo which blew my mind.

It is easy to start on Heroku, but staying there is a black hole in any budget.

Thankfully we've built Heroku Config Vars sync so one can move off of Heroku gradually.

>Heroku is unbelievably expensive.

only if developer/devops/admin time is free and for most it isnt. if 60 vs 250 /month is a breaking factor for you and your use case: go for it. if you need the software deployed to make real business it just doesnt matter.

$250/mo per process. And with appliku it is done automatically so there is no need for admin/devops time.

> if you need the software deployed to make real business it just doesnt matter

No wonder AWS can get away with the prices they have since people like you (and many more VC funded startups) simply have zero care about the price.

For people who run "real" businesses (meaning you need to have a larger income than expense [meaning, everything not VC-funded]), price is always a part of the calculations, even when it comes to hosting.

See https://news.ycombinator.com/item?id=28468943

To clarify, I mention "performance" in terms of the speed of "from commit to deploy", not the CPU/IO performance of an already deployed application.

That makes more sense, but the deploy cycle is already really fast with Heroku. I'm not sure if there is very much value in trying to shave off a few more seconds.

Depending on your requirements build ties in Heroku can be quite long. One app i’ve recently observed take 7 minutes to build because it always had to build both JS and Django parts(buildpacks for both were executed every time). We made it way faster with custom docker file that used caches and often built under 40 seconds. i am not saying that is not doable in Heroku, but would take more effort.

having something that specific, tailored for dynamic language, seems like overkill i'd undestand if it was for something highly performant and hard to deploy

Maybe, but maybe not. If you're using Python/JavaScript and have a ton of dependencies, most of your deploy-time is gonna be spent downloading dependencies. If the host you're using is specifically setup for Python/JavaScript, they can run a registry mirror with 100% of the packages right next to the host, and remove basically most time spent downloading the packages.

Same goes for other time-consuming activities like code coverage or even running the tests. If the applications running on the platform are somewhat heterogeneous, then you can start doing really aggressive performance work.

Reply on news.ycombinator.com