Hacker News Reader

Comment by jypepin | Hacker News Reader

You know what I think is missing on internet?

A good guide to transition from heroku, where everything is handled and you don't need to know anything, to AWS (or similar) where you end up having to either manage everything, or at least be aware of how things work and what are all the tools available (beanstalk, etc).

I find it missing, and it makes me scared of transitioning, while still be interested in learning more about server management / linux.

View on HN

I totally agree a good successor to Heroku is missing. A lot of organizations tried (Google AppEngine, Pivotal Cloud Foundry). I think AppEngine failed because the security sandbox was pretty constraining and Pivotal was too expensive for most companies.

With Kubernetes you can build the Platform as a Service (PaaS) on a shared Containers as a Service (CaaS) foundation (definitions are in https://github.com/cncf/wg-serverless/tree/master/whitepaper... )

You see multiple organizations trying this, Rancher 2.0 just got released, and we at GitLab just announced a partnership with Google's CaaS to make Auto DevOps more usable.

Our goal with Auto DevOps over Heroku is:

1. Open source 2. Run anywhere you want (don't pay extra for compute) 3. Smooth graduation (no need to start from scratch when you need a customization)

The biggest piece still missing to replace Heroku is idling of containers, we're hoping that this will get added to Kubernetes https://github.com/kubernetes/kubernetes/pull/29471

I think PaaS isn't really a thing anyone wants.

People want IaaS, FaaS or SaaS.

If you're smart enough to configure Elastic Beanstalk there isn't much holding you back from using CloudFormation.

I think there are a space between IaaS (request VMs) and FaaS (request functions). Current 12-factor apps don't run on FaaS and people rather not spend time setting up CI/CD, security testing, and monitoring for it.

BTW We're considering calling it an 'application platform' instead of PaaS, what do people think?

'Application platform' is a fine name. It gives a feeling of stuff being handled for you. However, 'PaaS' is an already established concept, so when I read 'application platform' I have to learn about what you mean with it. If you just say 'PaaS' then I can start thinking about whether it is good for me right away.

I really like where gitlab is going with regards to replacing many external services with integrated services within gitlab. Of course there is the concern of lock-in that is created this way, but I can't really pretend that teams don't get locked into Heroku, TravisCI/CircleCI, AWS, etc.

I've done it ~100 apps on heroku to internal PaaS. Reach out maybe we could make that guide :)

100 apps? That must have been expensive...

Mind if I dream too?

I want to see a react-native based full stack.

I feel like every app needs: css, html, javascript/react native, data management/redux, json, (LAMP is easy enough), PHP, larvel, mysql/mongo

Getting everything set up(IDE, servers, debuggers, and the code) for the first time has taken weeks, despite the actual programming being pretty simple.

Surprises me these dont exist since so many people have apps that require login and payments.

Have you seen

https://bullettrain.co/

Or something like https://aws.amazon.com/quickstart/ which is slightly lower level.

That might get you part of the way there. I suspect there's no business model for the full stack because every developer is going to say "it'll take a few days to set up and then we'll know the stack from first principles."

Bullet Train looks amazing, thank you!

I'd also love to see someone templatize backend stream processing and API integration systems.

> templatize backend stream processing

I may not get what you are saying, but isn't this what kinesis and kafka (which Heroku has an offering of: https://www.heroku.com/kafka ) do?

> API integration systems

So you're saying something like Zapier but more general? I'm not sure what you're looking for.

I took the classes even though I don't see a need to get the cert for my specific situation. The A Cloud Guru was indeed good.

I did watch them at 2x, though, and I skipped a couple of things... the specifics about how he setup WordPress hosting during my course was not how I'd handle it so I didn't bother with that part of it. I feel a lot better about my handle on AWS though.

On a thread yesterday, I saw someone post a link to convox. I could have sworn I saw a Heroku migration guide on their site, I even see one linked from their blog. But it just goes to the getting started section.

At any rate, convox looks to be pretty decent and perhaps might be nice for Heroku looking to migrate to something like AWS

Edit: link is convox.com

We run 25+ apps on convox. Let me know if you have any questions!

I have looked closely at deploying things to heroku and elastic beanstalk. I have looked at many many others, but these 2 appear the easiest and most battle tested. I had not looked at convox, though.

So my question would be, could you think of a reason to use convox over just elastic beanstalk? They both look relatively simple to me

When __would__ one exactly transition from Heroku to AWS? Any specific metrics to look at?

How much you're spending. :)

My point is more as an engineer, for the next app. I deploy things to heroku, simple enough, but when I feel constrained and want to learn a bit more to manage a little more things, where do I look?

To answer your question tho, a lot of things. Heroku makes it difficult to customize your stack, server, etc. Once your company and systems grow, you want more control.

Spend! I've encountered use cases where Heroku billing can be 5x what similar usage on AWS would cost. The difference between a $1000/mo AWS bill and a $5000/mo Heroku bill is justifiable when you're a small company that can't afford a dedicated DBA/Ops role, but it can break your margins when it grows to $25000/mo on Heroku vs. $5000/mo on AWS.

When __would__ one exactly transition from Heroku to AWS?

When they screw you on price/performance https://genius.com/James-somers-herokus-ugly-secret-annotate...

Wow, their rap genius platform with inline annotations and comments is actually one of the best blogging interfaces I've used. Blows medium out of the water.

There seems to be quite a few: https://www.google.com/search?q=migrate%20from%20heroku%20to...

Don't you think OP would have tried that.

Are any of those "good?"

Google App Engine has worked really well for me as a Heroku replacement. Check it out if you haven’t.

Almost everything seems to be missing this implicit change management.

Few tools grow with us users as we use them.

Why not?

just use lambda :D it’s a good in-between

I've heard good thing about lambda, but that's not the point of my comment.

TL;DR version:

1) Put existing system into maintenance mode (as much as possible).

  a) Write lots of tests

2) Build new system, copying verbatim (as much as possible). Resist temptation to improve.

3) Move functionality and/or adventurous customers over slowly.

   a) Fix bugs
   b) Repeat

4) Far in the future, move conservative customers over and retire the old system.

[This assumes your product is mature enough and not in active development, and can survive a dearth of new features for many months. If not, look into microservices and possibly move functionality over that way.]

Reply on news.ycombinator.com