Details
-
New Feature
-
Resolution: Won't Do
-
Major
-
None
-
None
-
None
Description
Given an application with a fixed number of processing threads, currently when a node becomes slow or fails, if those threads are switching between nodes in a gaussian way, they will ultimately all end up on the slow/down node impeding any requests to the still available/fast nodes.
When done, it should be possible to configure circuit breakers per node. For instance, given 100 'front end' threads and N nodes servicing operations, it should be possible to configure the system such that a maximum of 100/N requests are enqueued on a given node and the (100/N)+1th would fast fail automatically. In this way, even a multi-node failure would ensure data availability to remaining nodes.