I think I looked into this before and found that Metricbeat does not factor in the available value reported by statfs when it does its calculations. Filesystem pct/used not same as reported by df
So you think we should change the calculation in the Metricbeat code? If so we should also check the Windows code to see if it has a similar problem.