SWOT Heroku

Private Spaces, HIPAA, and several other features useful for anyone—not just big companies—require an enterprise agreement and sales calls. I would be surprised if there is a technical reason that is true, so this means that any company willing to sign a BAA without sales calls has potential to eat marketshare.

Imagine if Render creates a single-click private network solution and a signed SOC-2? Why would anyone want to have a phone call with Salesforce Enterprise Sales vs. clicking a few buttons in a UI?

This is a double-whammy for companies in a growth phase. It's extremely hard to predict infrastructure needs while growing, but the Heroku Enterprise process requires yearly commitments to do so (or it did they last time I interacted with it—how their Enterprise offering works at all is completely hidden).

Until you pay for performance dynos, which are overprovisioned for many use-cases, you won't have dedicated servers and you will hit "noisy neighbor" issues where your app inexplicably fails to perform for a brief amount of time. Diagnosing and eliminating noisy neighbors is hard and exists on any cloud platform.

This means that if you are designing your app to horizontally scale with standard dynos, you will forever be marred by this problem. Your alerting will always be noisy and your team will eventually ignore real issues because of alert fatigue.

Heroku suffers from first mover advantage, having built their own kinda-Docker-like-thing called buildpacks. While they have tried to get traction with that concept by open-sourcing it, Docker/containers currently dominate the rest of the world.

Heroku does support deployments with Docker, but it's extremely awkward to use. For each process type you must make a separate image, even if the only difference is the command line invocation used.

At Stitch Fix, we built a Docker-based deploy that used Procfiles the same way Heroku-without-Docker does. Our infra would schedule a given image, but use the Procfile to determine the command to run. The same image powers web and background jobs, as well as rake tasks and other one-off needs.

Why didn't Heroku build their Docker support this way?

Heroku Terraform templates don't let you do everything, and some don't even really work (you cannot upgrade datastores using Terraform). Their API is powerful, but Terraform is a very common standard for managing infrastructure and it sucks to be able to only use part of Heroku from Terraform.

Heroku Scheduler seems to have not been touched since 2013. It is not reliable and does not scale. There is no guarantee that your jobs will run on time or at all (and they don't always), and there is no way to observe its behavior or get errors. The web UI completely fails to scale because it truncates the commands and doesn't allow sorting. And you have to convert everything to UTC.

It feels very beta, and I think it should be retired if they aren't going to make it better. Everyone using Heroku should use their tech stack's cron-like service and just skip the scheduler. It's a shame because running scheduled jobs is hugely useful.

In the Common Runtime, all data stores (Postgres, Redis) are available over the Internet, protected by a password. These data stores do offer credential management and rotation (and they do so far better than competitors), but it does mean that if someone leaves your company, you have to rotate all the credentials of all datastores.

The only way to avoid this is to use the Private Spaces product which, as mentioned above, requires sales calls, price negotiation, and year-long contracts. I don't think this is the only technical solution to this problem, and I would bet that many teams would happily pay extra to ensure that only their dynos can access their datastores and not have them on the public Internet, even if such a setup isn't SOX or HIPAA compliant.

Heroku have added more metrics and monitoring over the years, but it's still in the vein of "useless information that makes nice graphs". We DO need to know when we are hitting memory and CPU boundaries so we know to scale or refactor, but the current offering isn't all that useful, especially for non-web processes.

To put it bluntly, you cannot use Heroku without also paying a third party APM or observability vendor. Why Heroku is happy to just let this money go straight to New Relic is beyond me. New Relic is not a good service, and Heroku could certainly build something of this level of quality and keep us on their platform.

This may sound like a minor annoyance, but when you have more than a few apps in your Heroku organization, it becomes very difficult to manage them. You have to make your own namespacing to ensure that your app names are available, since Heroku's apps are named from a shared global namespace.

If you are in the dark ages of using a staging environment, this, too, must be namespaces, so you end up with `acme-www` and `acme-www-staging` and run out of letters real quick.

Most developers think Heroku is toy. It is not. I met with a junior dev at a startup who was setting up fucking ECS to run his tiny app that was in beta and serving < 100 users. He thought Heroku was not a serious consideration. I told him that Stitch Fix almost went public on Heroku and that it would absolutely work for his use-case. I asked him to give a try. He later told me that after an hour he had migrated everything from AWS to Heroku…and was very happy about it.

This perception exists because Heroku either doesn't do any marketing or markets to…I guess legacy enterprise? I realize legacy enterprises have tons of money, but there is an entire generation of developers who will be leaders and decision makers in a few years who think Heroku is a toy. The fact is, a team can go from startup to going concern on Heroku and this is not something Heroku successfully promotes.

Why is there no chart on their website that explains the true cost of ownership of an AWS-based deployment pipeline? Why does the homepage say nothing about what Heroku actually is or does? Perhaps the startup market isn't large, but it is by definition growing. Successful startups become big companies. Unsuccesful ones mint engineers who will make decisions at their next job and you want them choosing Heroku.