6.6 Distributing files with Redis

When building distributed software and systems, it’s common to need to copy, distribute, or process data files on more than one machine. There are a few different common ways of doing this with existing tools. If we have a single server that will always have files to be distributed, it’s not uncommon to use NFS or Samba to mount a path or drive. If we have files whose contents change little by little, it’s also common to use a piece of software called Rsync to minimize the amount of data to be transferred between systems. Occasionally, when many copies need to be distributed among machines, a protocol called BitTorrent can be used to reduce the load on the server by partially distributing files to multiple machines, which then share their pieces among themselves.

Unfortunately, all of these methods have a significant setup cost and value that’s somewhat relative. NFS and Samba can work well, but both can have significant issues when network connections aren’t perfect (or even if they are perfect), due to the way both of these technologies are typically integrated with operating systems. Rsync is designed to handle intermittent connection issues, since each file or set of files can be partially transferred and resumed, but it suffers from needing to download complete files before processing can start, and requires interfacing our software with Rsync in order to fetch the files (which may or may not be a problem). And though BitTorrent is an amazing technology, it only really helps if we’re running into limits sending from our server, or if our network is underutilized. It also relies on interfacing our software with a BitTorrent client that may not be available on all platforms, and which may not have a convenient method to fetch files.

Each of the three methods described also require setup and maintenance of users, permissions, and/or servers. Because we already have Redis installed, running, and available, we’ll use Redis to distribute files instead. By using Redis, we bypass issues that some other software has: our client handles connection issues well, we can fetch the data directly with our clients, and we can start processing data immediately (no need to wait for an entire file).