I used to love, love, love Heroku. I would tell everyone about it, and at one point I was probably running 20+ projects on it.
I can't quite put my finger on why, but I've gradually transitioned everything off it over the past few years. I think it's been a combination of the ease of AWS, along with Heroku's changes to their Dyno pricing, but I never actually sat down and did the math. It just started to feel a little less controllable and a little more than I wanted to pay. To be fair, their database tiers were pretty pricey, so this article is nice news.
I'd be curious to hear if others have had a similar experience?
Their database prices are actually similar to AWS RDS Postgres now, and weren't that much higher before, especially with the nice features & interface you get in return.
Their dynos are insanely expensive though. A dedicated instance (VPS) with 2.5GB of RAM will set you back $250/month, while a similar EC2 instance will cost <$75/month.
It's hard to justify that the Heroku platform makes up for that $175/month/server difference compared to what AWS & co have nowadays.
I've always had this impression that AWS puts downward pricing pressure on itself, and periodically reduces its pricing, and that Heroku just swallows up the difference never passing it on to their customers. Would you say that is accurate?
We still use them pretty exclusively for our large projects. Their higher prices are still massively offset by not hiring a sysops team.
They have, however, put a big focus on their enterprise customers recently which has irritated me. That said, even after a few years, we still love them. They're leaders in PaaS, nobody else does as well as they do.
We use Heroku and only moved to it maybe two years ago. We have quite a spiky tragic pattern, both throughout the day and over the year - scale up 10x at Christmas. Heroku lets us do this super easily with no ops staff - we are a very small team.
The only problem we have is pricing for higher RAM dynos, we have quite a memory heavy app and need to use at least the 2.5gig dynos which are a massive jump in price. It’s not so much that we would invest in moving platform yet but it is something I look at every few months. However I do feel that Heroku are over due a price drop on their dynos (or just a ram boost at the same price point) and maybe a few more size options. They seem to be a little behind the market now - obviously a big part of their pricing is the ease of use...
It's because they haven't figured out how to keep the same level of experience when scaling past web+worker processes. As soon as you want to split your repo into multiple services the thing falls apart. They should add automatic internal service discovery (withing the same app), and ideally a way to map request path prefixes to different services.
That being said, Heroku+CDN (eg: Fly.io) is a fantastic combo for marketing websites. For $7.- / month you get a really nice development and deployment lifecycle. Especially the review apps matter the most to encourage good quality visual content.
We switched off Heroku to AWS Aurora once Aurora came out with a production-ready Postgres offering.
We haven’t looked back. On Heroku we used to run up against their CPU burstability limits and would get throttled. Simple Queries for an indexed field would take 100ms+. Web requests would queue up and customers would have a terrible experience.
The worst part is that Heroku support would often take days or weeks (!) to respond. And we were paying them $500+ per month. When they did, they gave us no insight into when we hit the burstability limits or when we’d be back to normal - the app simply got slow and it was never clear why.
Switching to aurora was fairly painless. And we pay less for a box that’s got roughly 4x the resources, decent management tools, and no burstability limits. It’s been fantastic.
We sent detailed and kind feedback to Heroku before and after switching and got a fairly bland response. Perhaps they couldn’t let us know that these kinds of new Postgres plans were coming, but it came off like Heroku didn’t really care much about the experience we had on their platform.
Have you experienced any pain-points with Aurora that perhaps you would have avoided by using RDS? We've been thinking of switching over, but some teams reported seeing different performance behavior from their existing queries that in some cases made them roll back.
Haven’t seen anything negative so far. It’s been a couple months now and super snappy. But perf with Heroku was so unpredictable that Aurora would have to be terrible for us to consider going back.
15 comments
[ 12.8 ms ] story [ 389 ms ] threadI can't quite put my finger on why, but I've gradually transitioned everything off it over the past few years. I think it's been a combination of the ease of AWS, along with Heroku's changes to their Dyno pricing, but I never actually sat down and did the math. It just started to feel a little less controllable and a little more than I wanted to pay. To be fair, their database tiers were pretty pricey, so this article is nice news.
I'd be curious to hear if others have had a similar experience?
Their dynos are insanely expensive though. A dedicated instance (VPS) with 2.5GB of RAM will set you back $250/month, while a similar EC2 instance will cost <$75/month.
It's hard to justify that the Heroku platform makes up for that $175/month/server difference compared to what AWS & co have nowadays.
Does anyone still uses Heroku?
To get actually more data on this, I’ll show a direct comparison – that same instance (Heroku Dyno Performance M) would cost you at
(Disclaimer: some of the services may differ in price due to different exchange rates and different taxation)AWS Fargate 1 VCPU, 3 GB RAM
Total = $63.89 + bandwidth (I believe dynos come with a generous bandwidth allocation included)we haven't migrated off yet, but something like k8s might be more attractive, as orchestration would be neat...
They have, however, put a big focus on their enterprise customers recently which has irritated me. That said, even after a few years, we still love them. They're leaders in PaaS, nobody else does as well as they do.
The only problem we have is pricing for higher RAM dynos, we have quite a memory heavy app and need to use at least the 2.5gig dynos which are a massive jump in price. It’s not so much that we would invest in moving platform yet but it is something I look at every few months. However I do feel that Heroku are over due a price drop on their dynos (or just a ram boost at the same price point) and maybe a few more size options. They seem to be a little behind the market now - obviously a big part of their pricing is the ease of use...
That being said, Heroku+CDN (eg: Fly.io) is a fantastic combo for marketing websites. For $7.- / month you get a really nice development and deployment lifecycle. Especially the review apps matter the most to encourage good quality visual content.
We haven’t looked back. On Heroku we used to run up against their CPU burstability limits and would get throttled. Simple Queries for an indexed field would take 100ms+. Web requests would queue up and customers would have a terrible experience.
The worst part is that Heroku support would often take days or weeks (!) to respond. And we were paying them $500+ per month. When they did, they gave us no insight into when we hit the burstability limits or when we’d be back to normal - the app simply got slow and it was never clear why.
Switching to aurora was fairly painless. And we pay less for a box that’s got roughly 4x the resources, decent management tools, and no burstability limits. It’s been fantastic.
We sent detailed and kind feedback to Heroku before and after switching and got a fairly bland response. Perhaps they couldn’t let us know that these kinds of new Postgres plans were coming, but it came off like Heroku didn’t really care much about the experience we had on their platform.