What I Learned Writing a Dropbox Clone - Part 4 - Conclusion

  commit f52479b
  Author: Rob Hoelz <rob@hoelz.ro>
  Date:   Sat Jan 3 16:33:56 2015 -0600
  
      Add notice about activity of this project
  

These posts are largely independent of each other, but if you'd like some context, you should probably read the first post.

For this final post, I'll talk about more “soft skill” type lessons I learned; lessons that seem to be common knowledge to many, but I still need to internalize in my day-to-day work.

Keep It Simple

The first thing I should have kept in mind is that the first release doesn't need to be perfect. My first iteration didn't have to address a lot of the more advanced things I thought of like decentralization, federation, perfect conflict handling, etc.

I often think “oh, it'll be harder to add that stuff in later”, recalling nightmarish systems I've worked on where internationalization or security were not a priority in the design of the system, only to become one later. But just because you're not building that feature into the system at the outset doesn't mean you can't plan the design to make adding it later easier.

I also relied too heavily on doing a lot of things myself, especially the hard things. I didn't use a framework for the server; I dealt with raw PSGI (which is Perl's analogue to Ruby's Rack or Python's WSGI). Instead of dealing with all of the quirks of inotify, I should probably just have used Git as a data store for the prototype. The inotify and transactional filesystem code was by far the hardest part of the project 1), and I should've waited until I could ride the rush of a release to refine it.

Release Early, Release Often

I worked on Sahara Sync for about a year before I gave up on it; release 0.01 (a typical first version number for Perl code on the CPAN) never saw the light of day. I had a working version of hostd about a month after I started; that would have been a good candidate for a first release to get feedback, interest, and perhaps some help in working on clientd.

Make Yourself Heard

I had plans to do this after 0.01, but I think I should have done it way earlier: I should have been talking Sahara Sync up. I didn't promote Sahara Sync at all; I don't even know if I talked much about working on it on social networks. If I'd promoted it, I may have gotten some interested developers involved to help me with the code, or at least provide moral support when I was feeling burnt out at the end.

You Have Reached the End of the Desert

Even though the project didn't go anywhere, I'm still proud of what I accomplished, and I learned a lot while working on it, both technical and non-technical. I'm happy with the overall design I created, and I think I had some really great ideas on what can all be done with a Dropbox-like system.

I've thought about picking up the project again, but in the last three years, the space has really changed: what was once a sparsely populated void has now exploded with projects. If I end up using one of them, I may see if some of the ideas I particularly liked about Sahara Sync would be compatible, and try to make a contribution upstream. Perhaps then, a bit of Sahara Sync would live on in a different way.

  1. 1) which was another reason the project was sunk: I had a Macbook at my job in Amsterdam, and I didn't have the energy to port that code to use FSEvents or kqueue