Who Needs One Database Anyway? byGlauber Costa

What if, instead of one database, an app created a million? This talk explores the trends driving this shift, why it makes sense for some use cases, and the infrastructure changes needed to support it. Get ready to rethink how databases scale.

1. A ScyllaDB Community
Who Needs One
Database Anyway?
Glauber Costa
CEO

2. Glauber Costa
■Former Linux Kernel contributor at Red Hat and others
■Former VP of Field Engineering at ScyllaDB
■Former Staff Engineer at Datadog
■Founder / CEO at Turso

3. ■Why multitenancy?
■How multitenancy?
■Architectural comparison.
Presentation Agenda

4. Why Multitenancy?

5. Bring in the law!
Shorter path to compliance
■Per-user isolation.
■Per-user encryption.

6. Developer Velocity
■Up if mistakes cost less.
■Shared resources increase
probability of mistakes.
■Shared database is source of terror.

7. Independent replication

8. Simplicity
■No need for row-level security.
■No need for caching (with replication).
■Independent backup / restores.

9. How Multitenancy?

10. Let’s do it!
■Pick a boundary that shards naturally (i.e. user)
■Make application aware of it.
■Every unit is just a database.

11. Data distribution drives cost
Expectation

12. Data distribution drives cost
Expectation Reality

13. Multitenant server:

Idle database costs 0 ( +
storage costs)

14. SQLite’s foundation:

Memory can be limited: SQLite pragma cache_size + heap limit functions

CPU can be reasonably limited: progress handler, how many bytecode
instructions are executed

15. Architectural Comparison

16. A familiar pattern?
●Consistent hashing?
●No sorting order across keys?
●Do you recognize this somehow?

17. Stay in Touch
Glauber Costa
[email protected]
@glcst
glommer
in/glommer