Heroku is up. My production application on Heroku is up.
Heroku's provisioning, build, and remote console services have been down for a few hours. Luckily my production app is still chugging along pretty well. There was a time this afternoon when I wanted to restart one of my six dynos (because it was half-dead) but was unable to due to the API lockdown. That meant every sixth request was sent to an app server that was likely to barf on it for a while there. The dyno got better a few minutes later though.
Postmortem edit:
API outage is over after a period of about three hours (that I was aware of anyway). During this window my app's response time was about 1500ms. Now that it's over it's down to about 250ms, which is on the high side of our daily variance between 150 and 250.
Unfortunately for us, we pushed out a marketing newsletter right about the time they shut down the API. Looks like we're still getting decent sales though!
Heroku is not accepting new builds, but is still performing for my web apps that are deployed there. I would say it's not down, but that they're having deployment issues.
Heroku isn't down, only some parts of the infrastructure are disabled. I haven't seen any service interruption to my production app however (which would be the case if Heroku as a whole was down, which is what the title implies).
I'm curious to know why it is that AWS can operate its API with a much higher degree of reliability than Heroku. They clearly have superior design and processes given the large difference in reliability.
Having been a heroku user for quite a while in the past I'm not surprised, these sorts of issues are sadly common with heroku.
What in particular is Heroku missing to pass level 3 and 4?
Hit and run comments aren't really useful. Some detail and if possible some suggestions for improvement really help. Even if they can't be implemented for some reason.
"We have completed the maintenance work. API functionality has been restored, and builds are resuming. We're continuing to monitor the affected systems." - Heroku Status
Builds are back up for me after a few hours of downtime.
I wish my company's IT team took their jobs as seriously and provided on-demand, easy-to-read timelines for service outages along with regular updates. That way anyone in the company can see at a glance how reliable the infrastructure is and how quickly issues are resolved.
Of course, the IT team would probably hate being held visibly accountable like that. But hey, I guess that's why they are working in corporate IT as opposed to a well-known and successful cloud service provider.
> Of course, the IT team would probably hate being held visibly accountable like that. But hey, I guess that's why they are working in corporate IT as opposed to a well-known and successful cloud service provider.
Is your IT team compensated at the same level as well-known and successful cloud service providers? If not, why would you hold them to the same standard?
Good point. Internal IT teams have a completely different set of incentives. Heroku's status timelines are important in keeping old customers as well as getting new ones. Internal IT doesn't really have that sales/retention problem, but they definitely have an "I might get fired if someone gets the wrong idea about an outage" problem.
We changed the title from "Heroku is Down" (originally, "Heroku has been down for more than 1 hour") in an attempt to make it more accurate. Happy to change it further if anyone suggests a better title.
In general, "Foo is Down" doesn't make for very good HN posts, because while it matters to users of Foo, the fact that Foo is down usually isn't intellectually interesting (which is what HN is looking for in a submission). Postmortems about why Foo went down, on the other hand, are often fascinating. Conclusion: we should usually wait for the postmortem.
Surely there's room on HN for original content in the form of comments? My top comment in this thread is a decent starting point for a discussion of what was actually down and what it meant to users. I'm sure Heroku's inside story will be even more interesting if they choose to blog it.
edit to add: Thanks for the transparency in moderation! I had thought this thread was poorly titled too.
Our website went from an appdex of .9 to .5, with page load times so long that many requests timed out. We're still pretty dead in the water for many users though it is getting better, according to NewRelic, and we didn't use Autoscaling.
Kind of surprised Heroku went down for so long in the middle of the day. Can't imagine any serious large scale services wanting to stay on a platform with that availability for very long. We're not large enough yet to move, but I think AWS is going to have to happen at some point.
I'm a long-time Heroku user (I've also written about them quite a bit, and published a book on the topic). In my opinion, Heroku is still, without a doubt, the best hosting option around for production apps.
I've worked and built enormous projects in the past and hosted them myself (on hardware, and on providers directly (Rackspace, AWS)), but have always had more headaches, wasted time, and downtime when doing things myself (and with a team) then when I'm using Heroku.
Regardless of the occasional incident Heroku has, I'm still 100% a loving user. The people over there work super hard on tons of stuff, and do a great job at keeping millions of applications up.
We're a Top-1000-ish site (in the US) with ~15M pageviews and ~50M backend transactions per month. We saw a ~20ms increase in response times across all requests (some are super-fast and skew things quite a bit) during the downtime. Luckily, our CDN strategy takes a lot of load of the backend and we were fine throughout the down time.
I'd echo rdegges point from eariler, Heroku has been (and will be for the foreseeable future) equivalent to a FTE in giving us the ability to build things and not worry (too much) about the hosting.
Edit: note the response time increase _during_ the downtime event.
24 comments
[ 2.6 ms ] story [ 55.5 ms ] threadHeroku's provisioning, build, and remote console services have been down for a few hours. Luckily my production app is still chugging along pretty well. There was a time this afternoon when I wanted to restart one of my six dynos (because it was half-dead) but was unable to due to the API lockdown. That meant every sixth request was sent to an app server that was likely to barf on it for a while there. The dyno got better a few minutes later though.
Postmortem edit:
API outage is over after a period of about three hours (that I was aware of anyway). During this window my app's response time was about 1500ms. Now that it's over it's down to about 250ms, which is on the high side of our daily variance between 150 and 250.
Unfortunately for us, we pushed out a marketing newsletter right about the time they shut down the API. Looks like we're still getting decent sales though!
[1] https://status.heroku.com/incidents/633
Having been a heroku user for quite a while in the past I'm not surprised, these sorts of issues are sadly common with heroku.
Heroku also has no transparency into its architecture, which is why no site on Heroku will be able to be PCI compliant after Jan 2015.
I'm not optimistic about Heroku's ability to continue to be a leading PaaS.
do you have a reference for that?
Heroku has never been able to pass a PCI Level 1 or Level 2 audit, but as of Jan 2014 it will also no longer pass level 3 or level 4 SAQs.
Hit and run comments aren't really useful. Some detail and if possible some suggestions for improvement really help. Even if they can't be implemented for some reason.
It's mostly documentation. If Heroku has built a secure system and documented it adequately, then it would easily pass SAQ A-EP.
One easy example: Maybe Heroku keeps logs of all HTTP requests that include params containing credit card numbers. Nobody knows.
Builds are back up for me after a few hours of downtime.
Of course, the IT team would probably hate being held visibly accountable like that. But hey, I guess that's why they are working in corporate IT as opposed to a well-known and successful cloud service provider.
Is your IT team compensated at the same level as well-known and successful cloud service providers? If not, why would you hold them to the same standard?
In general, "Foo is Down" doesn't make for very good HN posts, because while it matters to users of Foo, the fact that Foo is down usually isn't intellectually interesting (which is what HN is looking for in a submission). Postmortems about why Foo went down, on the other hand, are often fascinating. Conclusion: we should usually wait for the postmortem.
edit to add: Thanks for the transparency in moderation! I had thought this thread was poorly titled too.
Kind of surprised Heroku went down for so long in the middle of the day. Can't imagine any serious large scale services wanting to stay on a platform with that availability for very long. We're not large enough yet to move, but I think AWS is going to have to happen at some point.
I've worked and built enormous projects in the past and hosted them myself (on hardware, and on providers directly (Rackspace, AWS)), but have always had more headaches, wasted time, and downtime when doing things myself (and with a team) then when I'm using Heroku.
Regardless of the occasional incident Heroku has, I'm still 100% a loving user. The people over there work super hard on tons of stuff, and do a great job at keeping millions of applications up.
Keep at it! <3
I'd echo rdegges point from eariler, Heroku has been (and will be for the foreseeable future) equivalent to a FTE in giving us the ability to build things and not worry (too much) about the hosting.
Edit: note the response time increase _during_ the downtime event.