Durable Sync
⚡ Warning: The content of this page is obsolete and kept for archiving purposes.
Sync
- Sync started as an experiment in 2008 called “Weave”.
- It used to be completely under application services, but backend is now on our team.
- Does client side encryption.
- The password is really key here for the encryption key. You get a client side key.
- There are three sync clients in three languages:
- Firefox
- Fennec (Android)
- Firefox for IOS
- One of the problems here is:
- If you have a node go down and there’s not another synced devise up and running, you lose your data. Hence the focus on “data durability”.
- Adding a new data type is hard for Sync.
- Signing into Sync is complicated - ie, with Tokenserver, etc.
- Great diagram here on how existing version works.
- API docs live here.
Durable Sync
Overview
This is the internal name for the “new sync” project. Ephemeral data will become persistent data which in turn will (ideally) increase user’s trust in the system.
It’s a rewrite of an existing Python 2 service that was “bad for a number of reasons”. Most critical is data durability, ie data just gets lost. The goal here is to create a drop-in replacement for Sync. Same API, better/more stable backend.
It didn’t HAVE to be written in Rust, in fact the initial implementation/prototype was in python, but then it was decided to move forward with Rust.
Tokenserver
This points at various sync nodes given a specific key. Ie, you feed it the key, it knows which sync node to route you to.
It’s written in python2, and will be obsolete once the need for sharding goes away after the migration into GCP. Evidently we don’t need sharding for Spanner.
Detailed docs live here.