(no title)
tmcw | 2 years ago
Kysely is a little more established than Drizzle, which I think is one of the major reason why it has broader adoption. My bet is that Drizzle is moving really fast, gaining adoption, and might catch up at some point. It's also - in terms of performance - super fast, and nicely layers on top of fast database clients.
Some of the differences that I liked about Drizzle were the extra database drivers being core and developed as part of the main project. It supports prepared statements, which is awesome. The Drizzle API also covers an impressive percentage of what you can do in raw SQL, and when there's something missing, like a special column type, it's been pretty straightforward to add.
I prefer the way that it lets us write parts of queries, and compose them - like you import expressions like "and" and "eq" and you can write and(eq(users.id, 'x'), eq(users.name, 'Tom')) and you can actually stringify that to the SQL it generates. Or you can do a custom bit of SQL and use the names of table columns in that, like `COUNT(${users.name})`. I can't say scientifically that this is superior, and it's almost a little weird, but I've really found it a nice way to compose and debug queries.
That said, Kysely is also a great project and it'd be possible to build great products with it, too. I just found the momentum, API, and philosophy of Drizzle to be pretty compelling.
igalklebanov|2 years ago
Kysely is also super fast. Your bottleneck will always be database requests. If you're chasing every milli, why node.js?
> the extra database drivers being core and developed as part of the main project.
Kysely's dialects are dead simple to implement on your own. As evident by all the 3rd party dialects being open-sourced and all the comments from people using Kysely in production with stuff like cockroachdb, mariadb, clickhouse and such.
Its unhealthy to maintain niche database knowledge in the core. We just don't have the time (FYI we do this for fun, not trying to catch all the sponsors and get VC funded) to play around with all of these technologies, and stay up-to-date with changes.
Both Sami and I have submitted pull requests in 3rd party dialect repositories in the past. I maintain a few dialects on my own.
> It supports prepared statements, which is awesome.
In connection pooling scenarios Kysely was mainly built for, prepared statements are arguably "not that great". In FaaS, a burst of requests might make your database work extra hard, as each new lambda instance comes with brand new connection/s.
> I prefer the way that it lets us write parts of queries, and compose them - like you import expressions like "and" and "eq" and you can write and(eq(users.id, 'x'), eq(users.name, 'Tom')) and you can actually stringify that to the SQL it generates. Or you can do a custom bit of SQL and use the names of table columns in that, like `COUNT(${users.name})`. I can't say scientifically that this is superior, and it's almost a little weird, but I've really found it a nice way to compose and debug queries.
This has been part of Kysely for a while now, and is only getting stronger with new `ExpressionBuilder` capabilities. The fun part is, you don't have to import anything, and are not coupled to your migration code.
tmcw|2 years ago