Can hb_id always be unique?

[sidepelican]

Currently, hb_id is an integer sequence value in each process.
I don't fully understand why current hb_id' works like this, but for logging I would prefer it to be unique across all processes.

This creates extra work when I query logs associated with specific requests.
This is because logs are taken from each application server and put together (like in CloudWatch).

Vapor uses UUID, but Hummingbird is not.
Maybe this is because it doesn't depend on Foundation.
I think making a random string that's as long with UUID, with something like SystemRandomNumberGenerator, might be good.

[adam-fowler]

hb_id is generated from an incrementing 64 bit atomic. You would need to have 9223372036854775807 requests to your server before the id generated a duplicate.

[adam-fowler]

Or are you concerned about ids across multiple processes?

Rereading your issue it appears this is the case. I can look into changing this.

[sidepelican]

Yes, I'm assuming to run application in environments like ECS or Cloud Run.

[adam-fowler]

Just using a single 64 bit value random number will increase the likely hold of clashes. I might break this into two 64 bit values. A randomly generated number per server and the incrementing value.

And combine them into one id

[sidepelican]

UUIDv4 is 122 bit.
The proposed value is 128 bit but the half bits have bias.

I would prefer to have a completely random 128-bit value,
but it is possible that sequential values would have better performance than generating random numbers.

I think both are effective.

[adam-fowler]

#326

[sidepelican]

#326 seems to be merged. thanks!