Yeah, I have never *felt* the difference between native and electron more than when moving between Zed and VS Code.

His math is correct, but yeah he’s describing the behavior of the worst possible SOA design. Every half-good SOA plans for degraded states of the entire system when one service goes down rather than an outage.

Yeah, it’s too good to be true. To scale a graph DB you need to denormalize (or heavily index). Now you’ve reinvented Cassandra.

You might even call these services the “bedrock” of AWS

Bluesky uses Expo for web so it isn’t simple to add that feature right now, although soon Expo will introduce server rendering so it will be doable in theory.

Neon for Postgres, Turbopuffer for vector search, Warpstream for Kafka.

As cool as the query layer is, the expectation is that you run it with one of the supported storage layers like TiKV. These days I find more myself more interested in databases that are innovating at the storage layer themselves, such as those designed to write S3 for example.

In a sense, this is the most Bluesky of posts. (In a good way)

Scylla is great for things like social timelines which are inherently denormalized. Cockroach would be pretty expensive for the same use case.