The Architecture of Sharding

The Flow of Data From Request to Data

• The Client talks to mongos.
• mongos frequently talks to config servers.
• config servers contain metada about shard data, which data is stored in which shard. mongos queries these config servers often.
• mongos queries the shards for data.
• Mongo uses mongos to architect the routing from requests to shards.
• Multiple mongos processes can be up && running for higher availability.

About Shards and data

• Sharding is about splitting up collection data across "shards".
• Collection data is split across shards.
• Data in the shards moves around. config servers figure out which data has to move around between the shards to 'even out' the shard sizes.
• mongos can split chunks of data

Primary Shards

Sharded clusters have Primary Shards.
Each DB is assigned a primary shard.
Not all collections need to be, or are, sharded.
All non-sharded collections in the db are assigned to the primary shard.
the Config servers assign primary shards to dbs when they get created.
A DB's primary shard can be changed.