In our use case, we have an application or set of applications (websites, enterprise applications, and so on) that use the popular logback framework (http://logback.qos.ch) for logging structured messages to disk (access logs, errors, and so on). Currently, the only way to perform analytics on that data is to process the files in batches using something like Hadoop. The latency introduced by that process dramatically slows down our reaction time; patterns gleaned from the log data only emerge hours, sometimes days, after a particular event occurred and the opportunity to take responsive action has passed. It is much more desirable to be actively notified of patterns as they emerge, rather than after the fact.

This use case represents a common theme and has a broad range of applications across many business scenarios, including the following applications: