I'm interested in making a hot backup of my CouchDB 2.0 database onto an offline data store, such as S3, where I can't have a copy of CouchDB running.
To make a consistent copy of a small DB, I think I could issue a PUT to _replicator/backupJob with source on :5984 and target on :5986, and then copy the .couch file out of data/ (and onto S3) once the job disappears from the _replicator database. If I ever need to restore, then I could copy back from :5986 to :5984 to bring the backup online. However, there might not be space to do this for a multi-node DB. What is the best method to get a consistent snapshot? Even if I would take a set of nodes offline to cover each of the shards, and then grab the .couch file from each shard, I don't know how to ensure they are at the same sequence point. Does it actually matter if they are not quite at the same sequence point? (i.e. updates would go to the same shard) Christopher
