Connection Pooling and the Database Thundering Herd

1. Databases Fail on Connection Count, Not Data Size

Here is a failure that has taken down more startups than sharding ever has. The database does not fall apart because the data grew. It falls apart because the number of connections grew.

The two quantities move on completely different timescales, which is why one of them surprises people and the other does not.

• Data size grows over months and years. You watch it, you plan for it, and you have time to react.
• Connection count can double in thirty seconds, during a deploy, a retry storm, or a cache expiry