Split Mongo supports an arbitrary number of branches, though only "draft" and "published" are currently used. Course version history is preserved as edits are made, allowing rollback to a previous course version. Course locators for Split courses are the new style supported by opaque_keys. For example, here's a serialized Split course locator:

course-v1:SQU+SQU101+2014_T1+branch@published+version@8c056ceea2f35a1d705bd4c13d79c15b495a0f53

References - Old and New

Old

https://edx-wiki.atlassian.net/wiki/pages/viewpage.action?spaceKey=ENG&title=Mongostore+Data+Structure

https://github.com/edx/edx-platform/wiki/Split:-the-versioning,-structure-saving-DAO

https://github.com/edx/edx-platform/wiki/Split-mongo-architecture-and-rollout-options

New

http://edx.readthedocs.org/projects/edx-developer-guide/en/latest/modulestores/split-mongo.html

https://github.com/edx/edx-platform/blob/master/common/lib/xmodule/xmodule/modulestore/split_mongo/split.py#L2-54

Course Format in MongoDB

There are three Split Mongo collections in MongoDB:

• modulestore.active_versions
• modulestore.structures
• modulestore.definitions

active_versions

• Each course has a single active_versions document.
  • That document points to both a draft and published structures _id.

structures

• The entire course structure is stored in a single structures document.
  • Each time a structure changes, a new structure is created by cloning and changing the previous structure.
• The version of a course branch is the _id of the course branch's most recent structures document.
• Each course has a current draft structure and a current published structure.

definitions

• A course structure does not contain the course content. The content is kept in the definitions collection.

The Code

• Stored in common/lib/xmodule/xmodule/modulestore/split_mongo
  • https://github.com/edx/edx-platform/tree/master/common/lib/xmodule/xmodule/modulestore/split_mongo

Query Patterns

To retrieve the entire published course structure, the modulestore queries a single document from active_versions and another single document from structures (by _id).

Other Relevant Pages/Slides/Docs/Code

How to Recover from a Broken Split Course

A project was started to assist in determining information about the history of a Split course - and to rollback a course to an earlier version. It's here:

https://github.com/macdiesel/edx-split-utils

How To Find the Size of All Active Structures

Issue this command at the MongoDB command line of the replica DB:

db.modulestore.active_versions.find().forEach(
    function(obj) {
        var published_id = obj["versions"]["published-branch"];
        struct_obj = db.modulestore.structures.findOne(published_id);
        var curr = Object.bsonsize(struct_obj);
        print(curr + " :: " + obj["org"] + "/" + obj["course"] + "/" + obj["run"]);
    }
)

How To Walk Structure History Backwards From the Current Active Version 

rs.slaveOk();
var course_id = { "org": "HarvardX", "course": "SPU27x", "run": "2015_Q2"};
course_idx = db.modulestore.active_versions.findOne(course_id);
print("Course Index:")
printjson(course_idx);

var pub_struct_id = course_idx["versions"]["draft-branch"];

/* Walk the draft structure version tree backwards. */
var curr_struct = db.modulestore.structures.findOne( { "_id": pub_struct_id }, {_id: 1, edited_on: 1, previous_version: 1} );
print("Current draft structure: " + curr_struct._id);
print("Edited on: " + curr_struct.edited_on);

var generation = 1;
while ( curr_struct != null ) {
    curr_struct = db.modulestore.structures.findOne( { "_id": curr_struct.previous_version }, {_id: 1, edited_on: 1, previous_version: 1} );
    print("Previous: " + curr_struct._id + " - Edited on: " + curr_struct.edited_on + " - generation: " + generation);
    generation++;
}