User on Device with OS level encryption and SP2 backs up and restores Device

[rjcorwin]

Currently the backup for OS level encryption has no way of being restored without manual database work and it's also prone to crashes because it puts the entire database in memory. We should rearchitect this backup to write the changes feed to disk in batches and then write an import method in the UI.

[esurface]

Thanks RJ!

[chrisekelley]

We may wish to add proxy support for loading the dumpfiles in order to speed up the first replication; however, there are major caveats: Here is one issue that shows how it could (would?) throw an error:

pouchdb-community/pouchdb-load#54

The main reason to use the proxy flag is because normally pouchdb-load doesn't write any checkpoints. When doing normal replication after using pouchdb-load from dumpfiles, it will start reading all the changes from the remote CouchDB from the beginning of time, which could be slow. Using the proxy flag could enable checkpoints to be written and then replication would resume from the last checkpoint. This is pretty tricky though, because you have to include any filters used, and probably same goes for selectors, which are complex for pull. I'm concerned that since the docs don't even mention selectors may mean that the proxy code doesn't even know about them...

[chrisekelley]

The solution I've coded relies on 2 libs:

• https://github.com/pouchdb-community/pouchdb-dump-cli
• https://github.com/pouchdb-community/pouchdb-load

These are parts of the https://github.com/pouchdb-community/pouchdb-replication-stream ecosystem. We're using these parts rather than pouchdb-replication-stream because pouchdb-dump-cli enables splitting the dumpfiles.