Why is the _id a uuid

[singersbalm]

Why can't I define a custom _id?. I tried to define a users collection with the email as _id to improve lookup performance.

func isValidObjectId(id string) bool {
	_, err := uuid.FromString(id)
	return err == nil
}

Clover only allows uuids, but why. what's the benefit? Shouldn't that check be optional?

[ostafen]

Hey, @singersbalm, you are right, that check can be skipped.
Do you want to send a PR?

[singersbalm]

I would but I don't know with what to replace it. Should I replace the check with a string length check or should it check the env var for what kind of check it should use?

[mwmahlberg]

Forgive me, but this smells a bit like premature optimization. I have been running benchmarks, and unless you have several hundreds of thousands of users, the difference between using

_ := db.CreateIndex("users", "Email")
docs, _ := db.FindAll(query.NewQuery("users").Where(query.Field("Email").Eq("[email protected]"))

and accessing them via their ID field is negligible. Looking up users via index yielded the following results

$ go test -bench=. -v
=== RUN   TestBoltDB
--- PASS: TestBoltDB (0.03s)
goos: darwin
goarch: arm64
pkg: <does not matter>
BenchmarkBolt10kUsers
BenchmarkBolt10kUsers-10          250646              4753 ns/op
BenchmarkBolt100kUsers
BenchmarkBolt100kUsers-10         243933              5119 ns/op
BenchmarkBolt200kUsers
BenchmarkBolt200kUsers-10         265838              4894 ns/op
BenchmarkBolt300kUsers
BenchmarkBolt300kUsers-10         241802              4884 ns/op
PASS
ok      <does not matter>        658.752s

Mind you - that is lookups per second. Also note: Most of the 11 minutes was spent on creating the several hundred K users. The data model was simple, as we wanted to test the lookup, not the retrieval:

type User struct {
	Name  string
	Email string
}

Bottom line: It is highly unlikely that the lookup will have any significant impact when compared to actually loading the data.

From my point of view, using a non-customizable _id which is rather resilient against collisions and using indices to optimize queries is one of the key advantages of clover, keeping things simple. We could argue that for both bbolt and badger, using a BSON ObjectID would be an improvement, since it is monotonically increasing and hence better suited for prefix scans, but making the _id field configurable will open a whole slew of new problems.