It was definitely interesting to code this. It was 50% to meet a need, 50% to better understand Raft consensus in general, and Ben Johnson's go-raft implementation in particular (https://github.com/goraft/raft). As it happens Ben and I now work together, on InfluxDB (http://influxdb.com/), which has a new Raft implementation at its core.
Unfortunately Stripe doesn't seem to have kept the main site online, but the test harness is still available on Github.
What about the `.dump` command?
I've been using it to backup SQLite databases since forever. It produces an SQL script (similar to `mysqldump`) that you can import back into another SQLite database, or even into another RDBMS. It's much more elegant than copying the database file itself.
But come to think of it, I never really checked whether the `.dump` command produces a consistent backup, and the official documentation doesn't seem to answer that, either. If you're familiar with the internals of SQLite, please let me know whether I should keep trusting `.dump`.
I simply don't have enough transaction volume to test all the edge cases that might produce an inconsistent dump (and if I did, I wouldn't be using SQLite).
Regardless, this is a very interesting project, good job.
To be more precise, it's perfectly legal to use SQLite from multiple threads or processes at the same time. It will use its own mutexes and OS file-locking operations to ensure that at any given time, a database is being accessed by either a single writer, or an arbitrary number of readers.
In my opinion, the bigger issue is that clients can only safely access a SQLite database if it exists on a local filesystem. As soon as you decide you want to run multiple web servers, you're in trouble.
It saved me quite some hassle and reduced the attack surface compared to MySQL.
FWIW I never bumped into speed issues. My biggest issue against it is that it would let me alter columns directly which effectively sabotaged my use of alembic when I started using Python/flask.
The common thinking is that it's more of a replacement for fopen than it is for Oracle.
Any use of random(), as a trivial example.
Interestingly, once SQLite 4 comes out you'd probably implement this the opposite way, using Raft consensus (and something like etcd) for block storage, and using the SQL engine over arbitrary pluggable block storage.