MongoDB internals

MongoDB, a NOSQL Database designed to handle vast amount of load with horizontaly scaliablity

Why pop the Hood?

  • Understanding the data safety
  • Estimating RAM
  • Optimizing performace

Storage Layout

  • Every database is split into segments i.e single database is not just a single file instead, it is separated into multiple segments

Directory Layout


Creating a new Database journal

  • These files are the structure of the actual / physical data
  • At the time of the creation of new DB, if we don’t specify some special parameters the files/data allocates to us before we start using the DB.

    • Because allocating the file is quite expensive and for MongoDB to work properly, we take a file and set every character to 0.
    • .ns is a namespace file. a namespace is a collection or an index. a namespace is a collection of the database name and collection name ex: db.journal()
  • Thier is 1 namespace file per DB which can hold 18000 entries per default

    Tuning with Options

    • Use --drectoryperdb to separate DB’s into own folders which allow using different volumes (isolation, performance)
    • Use --smallfiles to keep data files smaller

      • If using thousands of collections & indexes, increase namespace capacity with --nsize

    Internal Structure


    DB does not store data in a continuous fashion in the file. It stored in extents i.e small patches of data. These extents are connected through each other via a linked-list . This is just because to make more manageable in Memory

    So, basically namespace has indexes that have a pointer that is pointing towards the file where data resides. In the file, we have different extents that store our document.

    Extent Architecture


    In extent there are several pointers , which we discuss

    1. Length: As its name suggest it is responsible for the containg the length of the extent
    2. xNext: Point to the next extent (as shown in the above photo)
    3. xPrev: Point to the previous extent , if is not there then points to null
    4. firstRecord: Point toward the first document in the extext
    5. lastRecord: Point toward the last document in the extent

    Linked-list of Linked-list that’s how Database is structure internally

    Extent and Records



    • Files on disk are boken into extents which contain the documents
    • A collection has one or more extents
    • Extent grow expoentially up to 2GB
    • Namespace entries in the ns file point to the first extent for that collection


    • Indexes are B-trees for structures serialized to disk

    They are stored in the same files as data but using own extents

    db.status() for Database statisitcs anddb.journal.stats() for Collection statisitcs

    Memory Mapping

    • All data filees are memory mapped to tVirtual memory by the OS
    • MongoDB just reads / writes to the RAM in the filesystem cache
    • OS take care of the rest

BSON (Binary JSON)

Data store in the document in the form of BSON i.e Binary-JSON representation . BSON extends the capabilities of JSON data-types

"name" : "Rahul Gupta",
"age" : 25,
"city" : "Delhi",
"county" : "India"

Data store in BSON havs field-value pairs i.e name → Field, Rahul Gupta → Value

  • The document can have a maximum of 16 MB so that it doesn’t use excesses amount of RAM for the operations.
  • If Document size is larger than MongoDB has GridFS for storing large document