Digging into a recent link for annotations on Google's SRE book led me to look back into the "Routergate"[1] issue in 2013.
For those who don't remember it, the short summary is that Heroku, a PAAS company of some repute, silently changed their routing for their customer's applications from least-used to random, without updating their documentation or their performance analysis tooling.
This had the following effects:
* Torpedoing the performance of any Rails applications with long-running threads
* Requiring many more workers (dynos in Heroku parlance) to achieve the same level of performance, at a cost of $35/month/dyno
* Inaccurate information was provided by the platform's integration with New Relic, an integration that adds another $40/month to the price of a dyno, making it hard for customers to see that their performance issues were the fault of the platform, not their app.
An info website was referenced, http://herokuclassaction.com, which is now suspended. The domain still resolves, but the old site was suspended by the webhost some time after February 2015 [2] and hasn't been usable since.
I've been unable to find much about the resolution of this controversy with a combination of google and HN searches.
Does anyone know what eventually happened?
Did Heroku ever pay Genius back the few hundred thousand they ripped them off with the silent change?
Did the lawsuit happen, or did it fizzle?
Did Heroku ever fix their routing to be less pathological?
[1]: http://www.wired.com/2013/03/hieroku/
[2]: http://web.archive.org/web/20150127094114/http://herokuclassaction.com/
Here's the rap genius blog post packed with links to replies and further comments.
http://genius.com/James-somers-herokus-ugly-secret-annotated
TLDR; They were upset that they had to discover the issue, eventually they changed web servers to use Unicorn and this fixed everything. Their final post on the matter was "It was all a dream".