The amount of VC dollars I have seen wasted on re-inventing the wheel of hosting/devops is a textbook case of premature optimization.
If anything, Salesforce has been moving more of its core developer experiences over to the Heroku model rather than trying to make Heroku more Salesforce-y.
SFMC is easily the worst Salesforce project from a dev experience / integration point of view, agreed there.
This would pretty much directly compete with Linode and DO on a value proposition. And a lock in to AWS ecosystem.
And they have a competitor to Linode/DO in their Lightsail offering (though how successful it is in competition I'm not sure)
I fail to understand this. If Heroku is on AWS, Heroku is selling its services at price X. AWS is getting from Heroku Y money, where X > Y (or it should at least, if Heroku is not operating at loss). If Amazon owns Heroku it could sell its services still for X, eating Heroku's margin plus the margin it already had on the AWS offering.
Parent was proposing that AWS sell Heroku services at cost.
As you say, my point was it doesn’t make sense
HeroKu Sell their services at (Y) where currently Y is >(X).
If they had acquired Heroku by Selling it at AWS Price (X), Amazon is still earning where they previously were not earning from those heavy Discount. At the expense of the Operational Cost of Heroku Team. And the benefits of further lock in and additional market segment which is currently not very well served by AWS.
I do wish Amazon had an offering like Heroku. While Elastic Beanstalk works pretty good it's nowhere near as nice or easy to use as Heroku. It has gotten a bit better with Amazon Linux 2 [Procfile, platform hooks], but still leaves a lot to be desired. I wish I could scale up/down faster like you can with Heroku. There is probably a way to improve it, but right now the way it's set up any deploy, config change, etc bundles the gems and precompiles the assets which take a while. So deploy/scale up take around 8-10 minutes. It's been a while since I've used Heroku but from what I remember it's pretty much instant to add more dynos.
I also ran into a couple of other annoying issues with EB.
1. There is a bug if using Amazon Linux 2 Ruby 2.7 deploying a RoR app where assets are not precompiled on configuration changes. It was an easy fix, just needed to add a script to .platform/confighooks/predeploy but it should have just worked.
2. When migrating to Amazon Linux 2 I deployed a version of the app that needed some code changes to work on Amazon Linux 2. There is no way to disable rollback on failed deploy on EB. Setting the "Ignore Health Check" option to true does not help here. I ended up in a situation where I needed to both push a code update and update some environment variables. However any code update I would push would fail because of the missing environment variables and rollback and any configuration change to update the environment variables would rollback because the current code had an error. I ended up having to tear down the environment and recreate it to get out of this loop.
3. Another minor issue is the Monitoring/Health dashboard. It will show warning/degraded/severe during scaling operations. For example when you trigger a scale down event it will complain:
"Environment health has transitioned from Ok to Warning. ELB processes are not healthy on 1 out of 6 instances. ELB health is failing or not available for 1 out of 6 instances."
Like of course it's not responding, the autoscaling group just terminated it.
I feel like Elastic Beanstalk really could be a Heroku competitor if AWS cared enough to put the work into it. It just needs some polish and quality of life improvements, but I guess this way they still make their money and Heroku has a good business middle-man'ing it.
I'm wondering if perhaps I would be better off using ECS or EKS, or if that's what AWS wants me to use hence why they are not putting much effort in EB.
There's a product opportunity for basically every AWS service to make it usable for those who just wanna get sh*t done.
This “git push” is a killer feature — maybe it doesn’t really make sense security-wise or in a real architecture, but it makes it very sexy.
I had a mainframes course in university and people from IBM would tout how everything we do today was already invented by them decades ago. But yeah, when you tried to use their stuff you had to jump through so many hoops, that you lost your motivation half the way.
My best guess is that a lot infra engineers actually value seeing the complexity and being able to understand all the pieces being used in the system more than they value getting something working quickly and easily.
Reminds me of the famous old HN comment about Dropbox
I’m not saying it’s right or wrong, but that’s where it comes from.
This post seems a bit disingenuous. Gitlap + Terraform + AWS != Heroku.
When it gets too expensive, Cloud66 is a great migration path and offers the best of both worlds. It's a fairly simply transition too.
Easy promotion of builds to staging/production after your tests pass and apps created automatically per pull request for testing in an environment just like all of your other environments.
Well that's annoying, seeing as I've just done everything else on the CLI
The quote was from the blog. Making it seem like doing things off the CLI was a plus point
Right now I have a git alias that does everything, including pushing to my heroku master, in one command. Why would I want to go off the CLI?
The "silly yml configs" are the things I care about the most. I don't want to have to remember how to reproduce my environment in order to stamp out a new one or recover an existing one. I'm all for removing the seldom-needed complexity, but "reproducibility" is the concern around which I build my whole tech stack--I largely avoid VM-based stacks precisely because the reproducibility story is so much flakier than the reproducibility story for containers. I imagine you can use something like Terraform for Heroku, but pretending that you can ignore reproducibility altogether is... really something. Maybe the parent means that "yaml is silly" and Heroku has a better reproducibility story in some other format (maybe graphical?)?
If you did all of the things required of launching a cloud level service for every signup page... few things would get done. There are some web apps which only need to exist for a few weeks or less before they can be tossed in a fire.
If I'm a startup with limited customer expectations of reliability and I just want to see if my friends would use my product demo... a point and click solution sounds pretty great.
"Update: This post does not live up to its original title We are building a better Heroku. It shows my own personal experience and reflects poorly on competitors. I am sorry about that."
The article doesn't make this case at all. Heroku is suitable for production and we didn't intent to suggest otherwise.
We’re trying to make the perfect production environment. Most people end up hosting at a hyper cloud (edit: like AWS). We’re trying to make the app on a hyper cloud in production as easy as Heroku is in development.
We’re still building that, it is t done by any means. And the article doesn’t do a good job of showing what we are trying to make easier. Setting up managed services on AWS is time consuming and that is what we want to solve. See https://gitlab.com/gitlab-org/5-minute-production-app/deploy...
The contents of the blog post doesn’t warrant the title. We regret it.
* The web app is deployed to a single AWS EC2 instance which cannot be scaled horizontally.
* The web server is deployed from an AMI filter, when a new AMI matches this filter there will be downtime to redeploy the instance entirely.
* There does not seem to be any considerations for patching the web server.
* Everything shares a single security group. Although this is probably fine because you are using managed services for redis and postgres its still weird to allow port 443/22 from anywhere on your database.
I'd be happy to be proven wrong on these points.
There also seems to be quite a bit of assumed knowledge about AWS and Terraform. You mention the free tier a lot, but that only applies for the first year since your AWS sign up. After the free tier is up this infra is going to cost $40+ per month.
FWIW, I've used Heroku as a production environment at multiple jobs. It works great there as well. The primary issue in my experience is cost. If you need to serve a lot of traffic, Heroku gets very expensive very fast. But it's still easy to use.
That cost can be offset by the fact that you may not need to hire a dedicated devops engineer right away, and your other developers spend much less time dealing with server setup and debugging.
I agree a simple production app can be great on Heroku. However besides being costly it is also hard when you want a managed service that isn’t on the platform. There are other reasons, https://gitlab.com/gitlab-org/5-minute-production-app/deploy... lists 9 in total.
While a production app on Heroku isn’t hard it isn’t trivial either. We’re trying to make the 5 minute production app faster to set up than Heroku for production apps by automatically configuring a bunch of services. We’re not there yet but that is the goal.
Why list all of these reasons and then make no use of them in the infrastructure code? The code doesn't scale, has huge downtime, you recommend granting Admin access to the CI jobs, the free tier is only for the first year the account exists. The purpose and focus of this repo isn't clear to me at all.
This code is awful.
> Terraform which has the following advanatges: > 1. Terraform is the most popular
What an awful reason to pick a technology. I love TF but using something because it's popular is the worst reason ever. Why even include this as a reason? None of the others are particularly compelling for a "5-minute" setup either.
> 5. You avoid the cost and complexity of Kubernetes
But k8s is popular so why not use that?
>If you are not sure what correct permissions are then use AdministratorAccess as a temporay solution.
Aside from all of the spelling errors in the README I hate this advice. You have the infrastructure code, you know what's going to be provisioned. Create an IAM policy and add it to the code or README. Telling people to give their little app Admin access is a terrible idea.
In addition to all of the issues another commenter mentioned. This is hardly "production" quality code or infrastructure. Production should be able to scale, that's the entire point of using HyperCloud providers.
They don't really mean "because it's popular", they mean because it's the industry incumbent and de-facto standard. You can't use anything else, because nothing else makes sense.
You need to standardize on one tool, especially if your infrastructure / teams grow. If you want to deploy something on AWS using an API with a minimal graph of dependencies, there's just Terraform. Pulumi isn't totally free, you have to write more code, there aren't nearly as many providers, not as many people know it, etc. There's basically nothing else. As a result, it's "popular".
"Setting up managed services on AWS is time consuming and that's what we want to solve."
Right, that's exactly what heroku has solved. Are you saying you don't think heroku has solved it as well as you have or you can?
> The contents of the blog post doesn’t warrant the title. We regret it.
OK, what SHOULD the title be, what's it actually about? Trying to build a heroku competitor?
Context: I have used Elastic Beanstalk in real production apps for years. For my Rails side project, figured I would use Beanstalk instead of Heroku so that I'm using the same platform (although I have also used Heroku in prod settings).
Quick summary is the AWS tutorial for launching their Rails sample app in Elastic Beanstalk doesn't work at all. I though maybe my Rails install was borked somehow, so I tried launching a sample app at Heroku. Worked on the first try.
Heroku really spends time making sure everything works out of the box, and that's what you're paying for.
And the next time I do it, it's like it's all brand new. It never gets easier.
Also it's very early stages - but the goal would to be better than Beanstalk
Click a button, bam deployed. Want to add a database, click what database you want, bam you got a database, environment variables set and a GUI (depending on what you pick). Want to scale up? Click a button, you're now scaled up. Want a entire review system with staging links and everything? Setup the review process line and you're done.
Everything is beyond easy and requires very little actual knowledge. You don't need a bunch of silly yml configs or knowledge of how containers work. You just pick a button and setup your git repo and it deploys your app. Want to SSH in? Install the CLI tools and type heroku ps:exec, want to see the logs? Go to the logs error to type heroku logs.
They have a perfect dev environment, fairly decent CLI tooling, bulletproof battle tested deployment setup (minus they fact they go down basically every 2 days if you aren't on the paid plans), and very little you need to understand.