Has anyone deployed Druid on Mesos, we would like to know your experiences or if this is not something recommended. Thanks.
I’m running our “small” cluster on mesos right now using marathon (dozens of nodes instead of hundreds).
Marathon has had a lot of changes to TASK_LOST handling in recent versions, and I’ve taken to using a 1.3 RC to get the latest changes. Mesos itself is running pretty darn well.
Make sure you understand the implications of the cgroups and especially memory limits with regards to memory mapped files.
As far as native mesos framework, there is no such thing at the moment, but such a thing that properly handles persistent storage is an interest of mine that I hope to eventually get to.
I don’t have auto-scaling working in the overlord yet (https://groups.google.com/d/msg/druid-development/1I3CmxlOipM/e3-SpWqG170J is the solution to handle it better, and is still in the works as a side project) but if your number of middle managers doesn’t scale, you can probably use marathon for that as well.