Though it closes the loophole for "free" apps which were keeping their dynos alive by being pinged frequently; now free apps can only be awake 18 hours a day, and go to sleep after 30 minutes inactivity.
But the $7/month hobby tier should make up for that, right?
You can just turn the pinging off. Then they're going to go to sleep by themselves. I've got a hobby app and it goes it sleep. The dyno comes back up in seconds. It's not bad.
Google App Engine also seems like a good place for a hobby app. More constrained, but Heroku is not really that much more flexible compared to totally free-form architecting.
Actually, looks like you could leave pinging on and Heroku will just force "exhausted" free dynos to sleep after 18 hours:
"During the beta period you’ll only get an informational email. In the future, we will force exhausted apps to sleep until they have been idle for at least six hours of the past twenty-four hours." -https://devcenter.heroku.com/articles/dyno-types#dyno-sleepi...
If you're looking for an alternative, you might want to check out https://www.tutum.co/. They are a bring-your-own-cloud solution that lets you deploy Docker containers to services like DigitalOcean and AWS, and they do a lot of the deployment automation Heroku does.
It's actually the same binary - the tutum agent just downloads it and upgrades it when requested. If you had docker already installed using a storage engine different than AUFS, because the tutum agent installer tries to use the AUFS storage engine (way more stable than "devicemapper"), you won't be able to "see" what you had in your previous storage engine. But the binary is the one from https://docs.docker.com/installation/binaries/#get-the-linux...
Our research suggests that almost all low traffic apps should continue to run fine on the `free` dynos. If yours don't, send me an email: pvh at heroku
Background workers aside, one of the most compelling reasons I've seen for a pinging service is the several seconds of waiting that beta users experience in experimental projects.
Running lots of experiments is always good (hobby or otherwise), and I'm very happy to see Heroku still embracing experimentation. But it hurts your experiment when you tell someone to visit a site and they click away after a few seconds of a blank page.
Is there any work going into speeding up the wakeup time?
I have had good experiences running several low traffic web apps (written in Clojure + Compojure + Clojurescript) on one Digital Ocean instance. Right now I am running a ton of stuff on one small Azure instance, getting perhaps just a few hundred thousand HTTP requests per day. But, a $5/month Digital Ocean VPS would probably handle all of your hobby sites.
I am looking to migrate a few of my hobby projects to the Heroku $7/month tier and try that for a few months as a test.
I am happy with their new pricing scheme - seems fair all around.
Really depends. Of course Heroku is way more convenient, but you can squeeze quite a few small apps on a $10 VPS. I'm not sure how they count ressource usage, but maybe an even smaller tier would be another option.
That said, for testing and stuff for personal use only the 18h deal is fine (although technically trolls could just go around calling random heroku instances every half hour, burning through the awake-time on free instances). And I can understand that they want to limit free usage.
And of course Redhat OpenShift still gives you three apps for free ;)
I had given some thoughts to this before. Using DigitalOcean (with perhaps Tutum) is cheaper if you are running a handful of small apps.
However the convenience of using something like Heroku can be underappreciated. I once lost data (docker data volume) on my Digital Ocean instance because I was too lazy to set up backups. With Heroku (and their Postgres service) I wouldn't have to worry about these "administration" stuff, and focus only the idea and the app ... which is exactly what I enjoy doing.
Even though I have good web engineering/devop skills I am thinking of using PaaS hosting even more than what I did in the past.
I was looking again at AppEngine (I used to use it a lot years ago, then stopped) but a recent AppEngine deployment of a small web app my daughter asked me for (to keep track of what books she has read, with some data import options) but I ran into a strange case where the beta search APIs worked fine in local dev mode and not in production. I spent some time tracking down the problem, then realized that PaaS hosting was supposed to SAVE me time.
Anyway, I have been waiting for Heroku's new pricing plan. I just set a hobby project to "free" mode and will deploy a few low traffic web apps to Heroku using the low cost $7/month tier and see if that fits my needs - definitely worth a few month test. I am mostly concerned that my $7/month (plus database) apps never get swapped out and the performance is good given that I may only have just several thousand requests a day (low traffic). I would expect Heroku to lower the resource priority on a $7/month app that used a lot of resources.
It sleeps after 30 minutes of inactivity, right? So no hits for 30 minutes means your app goes to sleep. Then the next hit will take longer while the dyno spins up before it returns.
Running 2 web dynos is the leanest recommended setting for production apps (obvious reason...). This means, in my case, a steep 45% increase for those small apps [from $34.50 to $50].
I'm not fond of this at all. Any small-scale apps (running 2-3 dynos) can see up to a 44% increase in price, and that's going to be very difficult to justify compared to the other PaaS offerings:
$100 a month on Heroku vs the previous $84.50 ($50 for 2 1X Dynos, $50 for Standard Postgres). $44.38 on BlueMix ($25.38 for 2 512MB Instances, $19.00 for 2GB/20 Connections on ElephantSQL Clearly not equivalent, but sufficient production-grade hosting for apps of this scale). Elastic Beanstalk hosting with RDS would be arguable even lower.
In a world where cloud hosting is becoming cheaper [0], I'm not sure how this jump this steep can be justified.
The parts things I'm pumped about:
- Paid hobby tier is brilliant idea for low traffic apps that require the guaranteed uptime
- Paid instances get analytics from the first dyno (I always found it absurd that I'm paying the same amount for two apps, but get analytics in one and not the other...)
- Worker's are now free on free apps, so it's going to be way less of a pain getting test and staging instances deployed
I'm also a big fan of the professional tier / PX dyno pricing. Brings it down from previously $0.80/hour, which is much more pricey than $500/month with their new pricing.
Still pretty expensive compared to taking the DIY approach on AWS, but I appreciate the price break here.
I imagine their markup on top of AWS is still something crazy like 50% or more?
Considering that m3.large (7.5 GB of RAM vs. PX 6GB) instances, on demand, are $0.14/hour, it's probably much more than 50%. (I'm sure they are using reserve pricing to reduce the price by 50-70%)
Cloud66 gives you most of the convenience of Heroku and lets you use your own cloud provider (AWS, Digital Ocean, etc)
Though it closes the loophole for "free" apps which were keeping their dynos alive by being pinged frequently; now free apps can only be awake 18 hours a day, and go to sleep after 30 minutes inactivity.
But the $7/month hobby tier should make up for that, right?