5 ms·

I don't often visit Heroku but it's always a trip to see their pricing. It's like the past 8 years of hardware cost reductions never happened, and you're still

by dahdum 10y ago

I don't often visit Heroku but it's always a trip to see their pricing. It's like the past 8 years of hardware cost reductions never happened, and you're still stuck paying $25/month for 512MB ram.

nkg 10y ago
What other platform would you recommend? I'm considering moving my next hobby app somewhere else...
- andrewstuart2 10y ago
  Move to an open-source orchestration platform and turn compute into a commodity so you don't feel tied to any particular VM vendor. I recommend Kubernetes, but there's also docker swarm, mesos, Nomad, Rancher, and probably about 100 other options out there, many of them very viable. In the very near future I really hope to see cloud vendors losing a lot of their business for hosted applications, and just becoming big elastic pools of bytes and FLOPs, with open source orchestration tools managing the consumption of those resources and keeping cognitive overhead low.
  - saool 10y ago
    I must be out of date, but my impression is that heroku's ease of use deploying and configuring applications through the web UI can't really be matched with kubernetes, mesos or others, which would need thousands of lines of orchestration code to work. Is that not the case anymore?
    - praneshp 10y ago
      Re Kubernetes, there will be some extra lines of course, but I really doubt it will run into the thousands. For example, I have a few (test) apps running on a (test) k8s cluster with a few small yaml files and a small shell script to build a docker image.
  - dahdum 10y ago
    Not sure if you're trolling or just love devops :). Since he said hobby app, I'd say throw it Digital Ocean for $5, running Ubuntu and letsencrypt for SSL. Use the DO guides to install whatever you want instead of paying separately for each service.
    - andrewstuart2 10y ago
      Haha yeah, just love devops I guess. Playing with kubernetes has been one of my hobbies for a few months. I suppose it's overkill for most hobby use cases, but I set up a small home cluster and have been loving it. It's much more natural now that I have some experience and now my little hobby apps can go on that cluster, and if I decide I want to deploy one for reals, it's just a few tweaks to get it going in any number of places.
      - dahdum 10y ago
        Right there with you...I setup a home Proxmox cluster for LXC / QEMU. Runs home automation apps, Plex, Gitlab, Rundeck, etc.. I like to call it Cloud Corner.
  - kdt 10y ago
    Obligatory "Heroku is Dead" article: https://circleci.com/blog/its-the-future/ https://circleci.com/blog/its-the-future/
- brianjking 10y ago
  Check out Flynn (https://flynn.io/ https://flynn.io/) & https://github.com/flynn/flynn https://github.com/flynn/flynn
- kkirsche 10y ago
  Honestly, google app engine especially if you can stay in their free tier
joshmn 10y ago
While I agree with your disdain, I have to remind you that there's a lot more to Heroku than just hardware. I mean, after all, they're sitting atop AWS. I can't go anywhere else do service create followed by git push service master and be ready to go within a minute.
- mderazon 10y ago
  You could go with Dokku and host it on a $5 Digital Ocean instance. Almost the same features plus you can install a Let's Encrypt plugin to handle SSL for free. Main difference is that you can't scale up - doesn't have that whole dynos thing. https://github.com/dokku/dokku https://github.com/dokku/dokku
  - jdietrich 10y ago
    The fully-loaded cost of a good engineer is $150-$300/hr. Unless you're operating at serious scale, the cost of Heroku is a rounding error.
  - lotyrin 10y ago
    How much would you pay a month, not to get paged when your single point of failure there fails, or you get traffic that outpaces it, or any of a number of other things? I'd pay more than 20 dollars.
  - joshmn 10y ago
    That's actually exactly what I do now for most of my projects. But it doesn't replace having to keep your OS updated, dangling with firewalls, etc.
- dahdum 10y ago
  You're right, I was a fan of them early on for the development ease and I still think they do a great job there. However the premium you pay for that experience continues to rise, while the alternatives are expanding dramatically (GC, AWS, DO, other PaaS).
- kkirsche 10y ago
lotyrin 10y ago
Still cheaper, easier and better than any team's infra I've seen to date. Usually there's some junk on DO or AWS, only a couple guys know where anything is. "Cheaper" because we don't track time lost to "what instance is <environment>.<app> on?" "How do I <dead simple task> without <toppling house of cards>?" Team A: had lots and lots of sites all on a single dedicated 2U rack server at a call-for-pricing provider, didn't really understand how to run it or that they had other options. Disk controller failed in a way that trashed the journal on the databases' volumes. They got a weekly backup restored but customers were furious. I had to explain that moving things to the "cloud" in the form of a single big VPS wasn't magically going to prevent similar risks. Team B: had 1 app, could have been a single dyno most of the time, scaled to 2 ahead of media announcements and the like (not worth considering/mo) and testing / staging environments could have been spun up with activity/on feature branches, so sub 100 bucks. Instead had 2 VPS (live/everything else) for only tens of bucks a month, but "dev-ops" guy never really completed or got assigned sprint tasks because he was too busy smelling like gunpowder and patching up holes in his feet. Team C: Big multi-tenant monolith-app on a dedicated rack at a call-for-pricing, wanted to figure out autoscaling, continuous deployment, review apps (every feature branch has an environment so you're not prematurely integrating code to get it deployed to a single bottle-neck staging branch+environment) how to bring up new apps (break tenants out to their own environments so they stop having noisy neighbor problems) etc so they could spend more time delivering value to the customer than having to ssh into things and running down checklists. They picked AWS, ansible, docker, kube -- everything "hip". Year later I think they're still "working on it". Team D: Enterprise, Everything is at central IT, we write apps with nice search features but IT doesn't want to learn how to spin up elastic search, so we get to write bastardized slow and less usable version. Meanwhile we're trying to figure out AWS, but everything's just pets managed by hand in the web console! We burn lots of money on people forgetting to shut down one-off instances and other people not having any info about what all of them are enough to clean up. Team E: "Knows what they are doing", hundreds of apps and services, owns their own top of the line hardware with a dedicated NOC team, those salaries actually amortize across each project pretty well, automates the hell out of everything with Puppet, doesn't bother with virtualization, they have the engineering resources to write code that manages real apps on real OSes on real hardware. However, most of that hardware sits with 50% RAM utilization and 5% CPU utilization. Day-to-day operations are smooth and very cheap but they way, way overspend on capital (hardware and automation code). Team F: Team E but nothing is automated or smooth, everything is virtualized and they spend too much on ops contractors getting paged at night to resize a 4gb LVM volume on a 20TB storage array (19TB free) because it got half-full of unrotated logs, or because a build box had two builds going at once and load average went above some silly arbitrary threshold. I'm a technologist but I hate watching tech organizations solve tech problems instead of business problems. I see lots of people doing the equivalent of drunk driving and complaining that Uber's so much more expensive than the gas. Of course, if you need and can build a software development lifecycle comparable to Heroku Pipelines or a PaaS comparable to Heroku itself or a database service comparable to Heroku Postgres, a IaaS comparable to AWS/GCE, etc. down to the metal and the cost of your solution plus the capital investment amortizes somewhere below what those services cost, go do that. Plenty of folks are in that position and should of course do so. Napkin math: It takes a whole year of dyno hours before you begin to approach what it costs to waste a day of a developer's time, or a 1 hour meeting between eight of them about "what branch is on <environment> right now?" It takes having a mean of 200 dynos running before you begin to approach what it costs to have a single ops salary.

by dahdum 10y ago

I don't often visit Heroku but it's always a trip to see their pricing. It's like the past 8 years of hardware cost reductions never happened, and you're still stuck paying $25/month for 512MB ram.

nkg 10y ago

What other platform would you recommend? I'm considering moving my next hobby app somewhere else...

andrewstuart2 10y ago
Move to an open-source orchestration platform and turn compute into a commodity so you don't feel tied to any particular VM vendor. I recommend Kubernetes, but there's also docker swarm, mesos, Nomad, Rancher, and probably about 100 other options out there, many of them very viable. In the very near future I really hope to see cloud vendors losing a lot of their business for hosted applications, and just becoming big elastic pools of bytes and FLOPs, with open source orchestration tools managing the consumption of those resources and keeping cognitive overhead low.
- saool 10y ago
  I must be out of date, but my impression is that heroku's ease of use deploying and configuring applications through the web UI can't really be matched with kubernetes, mesos or others, which would need thousands of lines of orchestration code to work. Is that not the case anymore?
  - praneshp 10y ago
    Re Kubernetes, there will be some extra lines of course, but I really doubt it will run into the thousands. For example, I have a few (test) apps running on a (test) k8s cluster with a few small yaml files and a small shell script to build a docker image.
- dahdum 10y ago
  Not sure if you're trolling or just love devops :). Since he said hobby app, I'd say throw it Digital Ocean for $5, running Ubuntu and letsencrypt for SSL. Use the DO guides to install whatever you want instead of paying separately for each service.
  - andrewstuart2 10y ago
    Haha yeah, just love devops I guess. Playing with kubernetes has been one of my hobbies for a few months. I suppose it's overkill for most hobby use cases, but I set up a small home cluster and have been loving it. It's much more natural now that I have some experience and now my little hobby apps can go on that cluster, and if I decide I want to deploy one for reals, it's just a few tweaks to get it going in any number of places.
    - dahdum 10y ago
      Right there with you...I setup a home Proxmox cluster for LXC / QEMU. Runs home automation apps, Plex, Gitlab, Rundeck, etc.. I like to call it Cloud Corner.
- kdt 10y ago
  Obligatory "Heroku is Dead" article: https://circleci.com/blog/its-the-future/ https://circleci.com/blog/its-the-future/
brianjking 10y ago
Check out Flynn (https://flynn.io/ https://flynn.io/) & https://github.com/flynn/flynn https://github.com/flynn/flynn
kkirsche 10y ago
Honestly, google app engine especially if you can stay in their free tier

joshmn 10y ago

While I agree with your disdain, I have to remind you that there's a lot more to Heroku than just hardware. I mean, after all, they're sitting atop AWS. I can't go anywhere else do service create followed by git push service master and be ready to go within a minute.

mderazon 10y ago
You could go with Dokku and host it on a $5 Digital Ocean instance. Almost the same features plus you can install a Let's Encrypt plugin to handle SSL for free. Main difference is that you can't scale up - doesn't have that whole dynos thing. https://github.com/dokku/dokku https://github.com/dokku/dokku
- jdietrich 10y ago
  The fully-loaded cost of a good engineer is $150-$300/hr. Unless you're operating at serious scale, the cost of Heroku is a rounding error.
- lotyrin 10y ago
  How much would you pay a month, not to get paged when your single point of failure there fails, or you get traffic that outpaces it, or any of a number of other things? I'd pay more than 20 dollars.
- joshmn 10y ago
  That's actually exactly what I do now for most of my projects. But it doesn't replace having to keep your OS updated, dangling with firewalls, etc.
dahdum 10y ago
You're right, I was a fan of them early on for the development ease and I still think they do a great job there. However the premium you pay for that experience continues to rise, while the alternatives are expanding dramatically (GC, AWS, DO, other PaaS).
kkirsche 10y ago

lotyrin 10y ago

Still cheaper, easier and better than any team's infra I've seen to date. Usually there's some junk on DO or AWS, only a couple guys know where anything is. "Cheaper" because we don't track time lost to "what instance is <environment>.<app> on?" "How do I <dead simple task> without <toppling house of cards>?" Team A: had lots and lots of sites all on a single dedicated 2U rack server at a call-for-pricing provider, didn't really understand how to run it or that they had other options. Disk controller failed in a way that trashed the journal on the databases' volumes. They got a weekly backup restored but customers were furious. I had to explain that moving things to the "cloud" in the form of a single big VPS wasn't magically going to prevent similar risks. Team B: had 1 app, could have been a single dyno most of the time, scaled to 2 ahead of media announcements and the like (not worth considering/mo) and testing / staging environments could have been spun up with activity/on feature branches, so sub 100 bucks. Instead had 2 VPS (live/everything else) for only tens of bucks a month, but "dev-ops" guy never really completed or got assigned sprint tasks because he was too busy smelling like gunpowder and patching up holes in his feet. Team C: Big multi-tenant monolith-app on a dedicated rack at a call-for-pricing, wanted to figure out autoscaling, continuous deployment, review apps (every feature branch has an environment so you're not prematurely integrating code to get it deployed to a single bottle-neck staging branch+environment) how to bring up new apps (break tenants out to their own environments so they stop having noisy neighbor problems) etc so they could spend more time delivering value to the customer than having to ssh into things and running down checklists. They picked AWS, ansible, docker, kube -- everything "hip". Year later I think they're still "working on it". Team D: Enterprise, Everything is at central IT, we write apps with nice search features but IT doesn't want to learn how to spin up elastic search, so we get to write bastardized slow and less usable version. Meanwhile we're trying to figure out AWS, but everything's just pets managed by hand in the web console! We burn lots of money on people forgetting to shut down one-off instances and other people not having any info about what all of them are enough to clean up. Team E: "Knows what they are doing", hundreds of apps and services, owns their own top of the line hardware with a dedicated NOC team, those salaries actually amortize across each project pretty well, automates the hell out of everything with Puppet, doesn't bother with virtualization, they have the engineering resources to write code that manages real apps on real OSes on real hardware. However, most of that hardware sits with 50% RAM utilization and 5% CPU utilization. Day-to-day operations are smooth and very cheap but they way, way overspend on capital (hardware and automation code). Team F: Team E but nothing is automated or smooth, everything is virtualized and they spend too much on ops contractors getting paged at night to resize a 4gb LVM volume on a 20TB storage array (19TB free) because it got half-full of unrotated logs, or because a build box had two builds going at once and load average went above some silly arbitrary threshold. I'm a technologist but I hate watching tech organizations solve tech problems instead of business problems. I see lots of people doing the equivalent of drunk driving and complaining that Uber's so much more expensive than the gas. Of course, if you need and can build a software development lifecycle comparable to Heroku Pipelines or a PaaS comparable to Heroku itself or a database service comparable to Heroku Postgres, a IaaS comparable to AWS/GCE, etc. down to the metal and the cost of your solution plus the capital investment amortizes somewhere below what those services cost, go do that. Plenty of folks are in that position and should of course do so. Napkin math: It takes a whole year of dyno hours before you begin to approach what it costs to waste a day of a developer's time, or a 1 hour meeting between eight of them about "what branch is on <environment> right now?" It takes having a mean of 200 dynos running before you begin to approach what it costs to have a single ops salary.