> SQLite isn't just on the same machine as your application, but actually built into your application process. When you put your data right next to your application, you can see per-query latency drop to 10-20 microseconds. That's micro, with a μ. A 50-100x improvement over an intra-region Postgres query.
This is the #1 reason my exuberant technical mind likes that we use SQLite for all the things. Latency is the exact reason you would have a problem scaling any large system in the first place. Forcing it all into one cache-coherent domain is a really good way to begin eliminating entire universes of bugs.
Do we all appreciate just how much more throughput you can get in the case described above? A 100x latency improvement doesn't translate directly into the same # of transactions per second, but its pretty damn close if your I/O subsystem is up to the task.
if you can tolerate eventual consistency and have the disk/ram on the application vms, then sure, keeping the data and the indices close to the code has the added benefit of keeping request latency down.
downside of course is the complexity added in synchronization, which is what they're tackling here.
personally i like the idea of per-tenant databases with something like this to scale out for each tenant. it encourages architectures that are more conducive for e2ee or procedures that allow for better guarantees around customer privacy than big central databases with a customer id column.
> personally i like the idea of per-tenant databases with something like this to scale out for each tenant.
So do I. And that type of architecture has come up a few times now in this comment thread. Given that Fly has the lead developer of the Phoenix web framework on staff, maybe it would make sense for him to work on integrating this type of architecture, with Litestream-based replication and the ability to have different master regions for different tenants, into Phoenix.
Not sure that would make sense. Phoenix uses the Erlang OTP platform, which ships with its own distributed database, Mnesia: https://www.erlang.org/doc/man/mnesia.html