When you tell Heroku to scale down its number of dynos it will still serve requests to dynos that are being shut down and it will kill dynos that are still actively serving requests. If you manually throttle dynos infrequently then you will probably not notice this, but the problem is still there. Since we were autoscaling based on live site usage and incoming requests, it expedited the visibility of this bug since our application would scale dynos up and down much more frequently to support application usage.
This bug has been confirmed by the Heroku support team and has not yet been fixed. When it is fixed I’ll post an update. In the mean time, scaling up is no problem, but if you scale down know that someone may get a Heroku HTTP 502 App Failed To Respond error because Heroku just killed a dyno serving a request.
blog comments powered by Disqus