Currently we report the error rate but if it is greater than zero it is hard for the user to find out what went wrong actually. As a first step:

• Rally will store more context about an error in its metrics store. This still requires that the user sets up a dedicated metrics store though.
• We will introduce a new command line flag --on-error which allows the user to either continue on error or abort.

This ensures that users can easily spot persistent errors by running a benchmark with --on-error=abort.