How do we assure a diffusion MRI processing pipeline that is: i) deterministic, i.e. given two runs of the pipeline on the same data, the same output is returned, ii) reproducible in time, and iii) efficient? Diffusion MRI has several processing steps that may not be reproducible between multiple runs. This reproducibility varies because of the parameters, multi-threading and the versions of the tools used. Moreoever, processing time for a large database can take several hours when each step are ran sequentially. To solve these problems, we developed a reproducible and efficient diffusion MRI pipeline based on Nextflow and Singularity.