Where I work today I have a complete copy of the database but not complete copies of all the files the system uses. Most of the time I can debug subtle problems from the production system quickly because I have all the data and if it has to do with recent data I can get a newer backup. When I work on features that need auxiliary files (images) I will go an download those individually for a few data records.

I’ve worked at other places where I could do the same and the volume of data was quite a burden which may or may not have been a problem, for instance when I worked on arXiv I had all the papers on my machine. Other times I have a reduced data set, when I worked on a patent search engine I needed to have about 10,000 documents to be able to train a good neural network but the full build for all the patents and non-patent documents took a day or two to build on a cluster.