Interesting read. I've been working on multi-device sync for some time and noticed a few things I ran into. The article speaks of P2P, and goes on to say that transaction logs are uploaded to the cloud. Not P2P then. P2P doesn't work anyway, because there's no guarantee that any two devices are active at any one time. I'm pretty sure my understanding of P2P is limited, compared to the author's.
The other issue is that all approaches described seem to preclude multi-user sync, which is what I've been struggling with. Security also becomes a ball-ache in that changing a password, which unlocks a private key that unlocks a symmetric key that encrypts and decrypts content means storing two copies of any one sync item until all sync clients have the new keys. There are ways of simplifying this, but complexity always creeps back in if I want a zero-knowledge server.
The third thing I'm working on is the idea of multiple, disposable servers. If one goes down, another steps in. If server A offers better comms to device 2 than device 1 which syncs via server B, then device 2 should default to server A without affecting sync scope from device 1.
We built a custom solution at SeatMe to sync multiple iOS devices with offline support to a central database. There were a lot of edge cases that make it pretty difficult.
There was the embedded mongo (or maybe couchbase?) db available for iOS that does multi-master sync that was being developed for a while that might work though.
I use Mongo to store entities. I front that with a relational database to a. map entity locations and b. hold associations. That lets me track entities by device/application/users (one item can be shared bu multiple apps and multiple users, each using multiple devices). I want to be able to disable devices and user accounts. Only the creator of an item can hard delete an item, and so on. My code is custom too, because there's nothing out there I could find that does that.
I already use sqlite in the device (postgres in the server) and have the app almost finished, so I wish to not switch datastore, specially one that is not relational/transactional.
This is great. Guys, I love what you're doing with objc.io. Please keep it up!
Everyone else: if you want to help support the site, set up a recurring purchase for the objc.io app in the App Store. It's something ridiculously cheap (for the value provided), like $5.
Also, for more on iOS data syncing and Core Data, this has been a frequent topic of late on Brent Simmons' blog: http://inessential.com
The other issue is that all approaches described seem to preclude multi-user sync, which is what I've been struggling with. Security also becomes a ball-ache in that changing a password, which unlocks a private key that unlocks a symmetric key that encrypts and decrypts content means storing two copies of any one sync item until all sync clients have the new keys. There are ways of simplifying this, but complexity always creeps back in if I want a zero-knowledge server.
The third thing I'm working on is the idea of multiple, disposable servers. If one goes down, another steps in. If server A offers better comms to device 2 than device 1 which syncs via server B, then device 2 should default to server A without affecting sync scope from device 1.
It's a fascinating subject.
Good article, all the same!