[evanelias]

Other times, they would directly talk to a single employee, but get skewed or misleading information based entirely just on that one employee's POV.

Their post about Tumblr's architecture [1] focused a lot about JVM-based services, HBase, etc which in reality was only ever used for a tiny subset of the backend. The huge section on "Cell Design for Dashboard Inbox" was especially ridiculous: the systems described there were literally a mix of complete vaporware and failed/canceled projects that never even got close to production.

As an early Tumblr engineer, I was really upset to read this nonsense. I spent several months of my life working very long hours to successfully scale the existing (PHP/MySQL) dashboard activity feed architecture in 2011-2012. It continued to be used as-is for many years after this interview, with lower latency and much lower cost than the proposed hbase/scala cell replacement.

And of course, engineering candidates being interviewed would always ask about this hbase cell architecture thing that they read about in High Scalability...

[1] http://highscalability.com/blog/2012/2/13/tumblr-architectur...