[barking_biscuit]

What is considered a "very low write throughput"? Are we talking 10s of transactions per second? 100s of transactions per section? 1000s of transactions per second? 10,000s of transactions per second?

https://stackoverflow.com/questions/35804884/sqlite-concurre... has some pretty interesting numbers in it. Particularly the difference between using WAL mode vs not. The other aspect to consider is that the concurrent writes limitations apply per database file. Presumably having a database file per tenant improves aggregate concurrent write throughput across tenants? Though at the end of the day the raw disk itself will become a bottleneck, and at least the sqlite client library I'm using does state that it can interact with many different database files concurrently, but that it does have limitations to around ~150 database files. Again, depending on the architecture you can queue up writes so they aren't happening concurrently. I'd be very interested to know given all the possible tools/techniques you could throw at it just how far you could truly push it and would that be woefully underpowered for a typical OLTP system, or would it absolutely blow people's minds at how much it could actually handle?

When it comes to SQLite performance claims I tend to see two categories of people that make quite different claims. One tends to be people who used it without knowing its quirks, got terrible performance out of it as a result, and then just wrote it off after that. The other tends to be people who for some reason or other learned the proper incantations required to make it perform at it's maximum throughput and were generally satisfied with its performance. The latter group appear to be people who initially made most of the same mistakes as the former, but simply stuck it out long enough to figure it out like this guy: https://www.youtube.com/watch?v=j7WnQhwBwqA&ab_channel=Xamar...

I'm building an app using SQLite at the moment, but am not quite up to the point in the process where I've got the time to spent swapping it out for Postgres and then benchmarking the two, but I dare say I will, as I have a hard time trusting a lot of the claims people make about SQLite vs Postgres and I'm mainly doing it as a learning exercise.