Documents are stored in natural order
The documents are stored in descending order based on date. So the collection has 20140731 as the first document.
Unless you are using a capped collection, there is no guarantee for the ordering of documents on disk (also referred to as natural order).
Document deletions and moves (when a document outgrows its allocated record space) create space on the free list which will be reused.
Here's a quick example which should demonstrate this in the mongo shell:
// Start with an empty database & collection
use demodb; db.dropDatabase(); db.order.drop()
// Add some test data
for (i=0; i<1000; i++) {
db.order.insert({'i': i})
}
// Looks like insertion order! (0..9)
db.order.find({}).limit(10);
// Pause 5s for effect :)
sleep(5000);
// Remove half the entries
db.order.remove({ i: { $lt: 500 }})
// Re-add the missing entries
for (i=0; i<500; i++) {
db.order.insert({'i': i})
}
// Not the entries you expected .. space from deleted records was reused
db.order.find({}).limit(10)
// Clean up demodb
db.dropDatabase()
Order of results
When I use the find command with filters {$gte : 20140720, $lte : 20140731}, mongodb returns back the query in ascending order of "date" field.
If an index is used for a query, the documents are returned in the order they are found in the index. You should take advantage of this when constructing your indexes for common queries (see: Use Indexes to Sort Query Results).
FYI, a simple index (eg. on {date:1}) can be used to return the results sorted in either ascending or descending order.
Sort by ObjectID
If you are using MongoDB's default ObjectIDs for _id, you can sort by { _id: 1 } to approximate insertion order since the first 4 bytes of the ObjectID incorporate a timestamp. If you wanted to use this for sorting a query based on date and approximate insertion order you would ensure an index on {date:1, _id:1}.
Note that ObjectIDs are typically generated by the client driver, so if you have clock drift on your app servers (or the _id is created some time before the document is inserted) the ObjectIDs may not strictly reflect "insertion order" as seen by the server. If accuracy of insertion order is highly important, it is generally possible to generate the _id on the server-side (approach varies depending on the driver).
