Brandur I think you left just before we started to have real headcount problems which I blame for our demise. I agree with pretty much everything you said but I think the technical and product problems you bring up were actually symptoms, not causes.
A few more people and we could've properly maintained the stacks. We should've had an entire team if not department dedicated to thinking about Docker. If I recall it was a part-time concern for like 2 of the language engineers (who had a ton of work on their plates).
Heroku should not have stopped at Cedar: it should've been the one to build something akin to k8s but with good DevEx.
As far as why we had headcount problems: it's because sfdc gave up on us. See Craig[0] and Adam[1]'s comments for more on that. We certainly did though: sfdc would have entire teams dedicated to similar roles we would have individuals handling part-time. Salesforce folks would tell you we were cocky assholes and they were right—but we also worked damn hard.
Rightfully so. They’ve been completely silent on the GitHub integration and when it might be restored. It’s also extremely difficult to migrate off of Heroku. It’s either stay in Heroku and hope it doesn’t implode further, or higher a DevOps person to wrangle K8s on GCP or AWS. It’s a nightmare and there really aren’t any alternatives to Heroku (I don’t care what anyone says about render or fly. They aren’t there yet).
Depending on your specific requirements, take a look at AWS Elastic Beanstalk, or GCP Cloud Run (some glue required for the latter). If you do end up going the DIY route, maybe take a look at Nomad before k8s, it's drastically easier. Disclaimer: i work at HashiCorp, but i've been blogging about this for years before joining (blog link in bio).
It's odd how whenever someone on HN mentions a list of cloud providers, they will list Google, IBM, and even Oracle before Microsoft Azure, which is the #2 biggest provider after Amazon.
It's like when a parent lists their children and just "forgets" their second child.
To be fair, Azure has a bad reputation among non-"enterprise" folks ( the type of person who swears by IBM, Oracle, Cisco, etc.). I've only experimented with it briefly and it was underwhelming at best, and all of the things I've read and heard about it don't make me want it. Be it general slowness, really poor UI and UX, or terrible security vulnerabilities. No cloud is perfect, but Azure seems to be the least perfect of the big ones.
Totally agree that it's strange. I hate Microsoft MORE than the next guy. I will never get over the company they were in the 90's, and funding the SCO lawsuit against IBM. And I view the warm, fuzzy "Linux-friendly" Microsoft with high-levels of skepticism. That all being said, I have a couple of larg-ish, cross-departmental apps running on Azure infra, and I find their portal to be very good; a mile ahead of futzing around in AWS. (Of course, I understand that AWS was designed to be CLI-friendly first, but I don't need to deploy fleets of similar objects at a time.) So I can't fault Microsoft for Azure. I actually like it. One of my apps is a Rails app (with all the attendant nonsense) and one is the backend of a Windows-native app, and the integration with Visual Studio for deploying services to their cloud really is nice.
I should have been more clear. It’s not impossible because the physical act of migrating. It’s impossible because so many other platforms are missing tablestakes features (HA Postgres with auto backups as an example).
React SPA front end. Hasura for GraphQL. Distributed Elixir (which Render handles brilliantly and Heroku is not even capable of). Also an old Rails app with web server & background job server.
I use RDS on AWS because I'm doing some weird stuff that Render hosted Postgres can't handle, but nor can Heroku Postgres.
I have never used Heroku so I have no dog in this fight but it oddly seems like a weird smear campaign, esp accompanied by threads promoting it's competitors and all these weird brand new 'throwaway' accounts posting in these threads. I get that they have an ongoing issue but a month ago Jira had a multiple weeks long outage and it basically only got like 1-2 threads total that made it to the front page that I saw (one about the initial outage and one the postmortem).
The ongoing security incident has caused a cascading series of former Heroku employees to come out sharing their perspective. Not too strange given the circumstances.
> One of the things Patrick would assiduously list as an existential threat to the success of Stripe was a major security incident
When I worked at AWS, the thing I heard most frequently repeated in relation to security was "We don't want a front page of the NYT event." There's probably a good argument that AWS or Amazon would survive a "front page of the NYT" event (after all, we've seen Target etc do so), but the spirit of the phrase still resonates.
This is one of the main things we cover in all of our security awareness training. Yes, breaching data is worst case scenario and can be devastating to individual users, but it’s often the brand that takes the hardest hit.
You’re missing the point. It’s not to show that the brand is more important than user data, it’s to show that security events are multifaceted. You want employees to avoid making mistakes that lead to security events. Showing all the ways those events can fuck things up is the point.
Well, all problems are multifaceted. If I had to take trainings about how not to spill toxic waste in local rivers, I would not find it appropriate to mention damage to the brand either even though the brand certainly would suffer from such an event.
I think inherently we're all pretty selfish (for better or worse). I'd guess that saying "we can see the from the past that the brand is damaged enough to cause XX% layoffs/go under" - as in an implicit 'this kind of thing will have real life consequences to your job' - probably has more impact than saying "there are real world consequences to people you have and are never likely to meet". Not saying it's right, but it's probably true.
> Selling to Salesforce for $212 million was an obvious win
NO! I always thought the very opposite.
Selling Heroku to Salesforce was a mistake.
Heroku should have cannibalized it's own dyno pricing and follow AWS Lambda with 100ms increments pricing.
And after that do what Cloudflare is doing today.
Selling it to anyone will be difficult. Way too much legacy, and lost context. It would be a quagmire of inertia to shift anything. As with most tech "giants"/"unicorns" that are past their zenith.
Most likely any buyers would kill off/sunset most of it, and gain a lot of negativity. Not going to happen.
More likely is the rumours of it being rolled/rebranded/de-labelled into a Salesforce offering, and quietly sunset large chunks of it (which would be less noticeable than if any new buyers did so).
A few more people and we could've properly maintained the stacks. We should've had an entire team if not department dedicated to thinking about Docker. If I recall it was a part-time concern for like 2 of the language engineers (who had a ton of work on their plates).
Heroku should not have stopped at Cedar: it should've been the one to build something akin to k8s but with good DevEx.
As far as why we had headcount problems: it's because sfdc gave up on us. See Craig[0] and Adam[1]'s comments for more on that. We certainly did though: sfdc would have entire teams dedicated to similar roles we would have individuals handling part-time. Salesforce folks would tell you we were cocky assholes and they were right—but we also worked damn hard.
[0]: https://news.ycombinator.com/item?id=31373394
[1]: https://news.ycombinator.com/item?id=31373300