How Discord Stores Trillions Of Messages

Discord’s journey from MongoDB → Cassandra → ScyllaDB highlights the challenges of scaling a high-volume messaging platform.

1. Initial Stage (2015) — MongoDB

Database Used: MongoDB

Scale: 100 million messages

Challenges:

• Initially, MongoDB was sufficient for Discord’s message storage.
• However, as the volume of messages grew, data no longer fit in RAM.
• Latency became unpredictable, making it unsuitable for a large-scale messaging system.
• MongoDB wasn’t scalable enough for Discord’s rapid growth.

Decision: Move storage from MongoDB to another database.

Chosen Solution: Apache Cassandra.

2. Scaling to Billions of Messages (2017) — Cassandra

Database Used: Apache Cassandra

Scale: Billions of messages

Challenges:

• By early 2022, Discord had 177 nodes storing trillions of messages.
• Latency became unpredictable, and maintenance operations were costly.

Key Issues:

• Cassandra’s LSM Tree Structure…