sqlite

> You can get substantially better performance out of sqlite by using the lower level https://github.com/crawshaw/sqlite, turning on WAL etc, using a connection per goroutine for reads, and sending batches of writes over a buffered channel / queue to a dedicated writer thread. That way you can turn off SQLite’s built in per-connection mutex but still be thread safe since each connection is only used on a single thread at a time. Would this protect against a row update in the middle of a read? e.g. would a row at least be internally consistent at the time it was read?

I’m not sure to what degree you want to follow the Redis no concurrency “everything serialized on one thread” model. You can get substantially better performance out of sqlite by using the lower level https://github.com/crawshaw/sqlite, turning on WAL etc, using a connection per goroutine for reads, and sending batches of writes over a buffered channel / queue to a dedicated writer thread. That way you can turn off SQLite’s built in per-connection mutex but still be thread safe since each connection is only used on a single thread at a time. For this use-case you will also probably save a lot of time if you use some large arena-style buffers (probably N per conn?) and copy incoming parameter bytes from the network request/socket to the buffer, or copy straight from sqlite out to the socket, instead of allocating and passing around a bunch