How does sharding differ from replication?

Sharding and replication solve different problems, even though both relate to distributed data in MongoDB.

#Replication

Goal: fault tolerance and read scaling. The idea: the same data is copied onto several nodes.

• there's a Primary and several Secondaries
• every node holds the same set of data
• if Primary fails, a Secondary takes over
• reads can come from Secondary, taking load off Primary

The key idea: several copies for reliability and availability.

#Sharding

Goal: scaling by data volume and write load. The idea: data is split into pieces (shards), and each piece lives on its own server.

• each shard holds only part of the collection
• load is spread across different nodes
• the database can grow almost without limit
• writes and reads are distributed across shards

The key idea: splitting large data across several machines.

#In one sentence

• Replication, the same data on several servers (reliability + read scaling).
• Sharding, different data on different servers (write and volume scaling).

If you'd like, I can give a simple analogy in my next message, something like "replication is like photocopies, while sharding is like chapters of a book kept in different rooms".