We have a few very large Matrix Configuration project used to distribute work out to a cloud of Hudson slaves.

The overall job takes 4+ hours to run and uses many external resources that could become unavailable briefly through the run. Any failure causes that specific axe to fail and correctly rolls up the entire job as failed after the run is complete.

After completion we need to re-run to hopefully fix the axes that failed, but have no interest in re-running the axes that were successful!

An option to only queue the axes that failed on the previous run (or even better any run that is in the job history) would be really awesome, saving countless hours wasted by axes that must start up only to realize they have no additional work to perform.

Thanks for your contributions to Hudson!