Collisions of Ids about newid HOT 8 CLOSED

If it doesn't work for your use case, don't use it.

The goal was not just unique, but ordered identifiers.

from newid.

Hi @phatboyg,

I'm only stating that fact that GUIDs should guarantee uniqueness. Order and sequence can still be maintained with a high degree of quasi-uniqueness using a different and much faster technique. This is not just in my case, this is everyone's best interest. GUIDs have the assumption of being unique.

You just mentioned that:

The goal was not just unique, but ordered identifiers.

The order part your is maintained, but the most important part is uniqueness, which is not guaranteed with your current implementation. I think you should mention this limitation in your docs. Users of this library may not be aware of this.

Anyways, I'll close this issue.

from newid.

That's just it though, they are unique. If you are seeing a case where they aren't, I'd like to know about it.

And I'm going to add a method to create multiple ids in a single call, which basically cuts the time down significantly to generate 1,000,000 identifiers. So if you ask for 10, you can get 10. Or maybe I can keep an internal buffer and just allocate them more quickly. Maybe. Hmm.

from newid.

Hi @phatboyg,

Here's the code sample that was used to show collisions.

    static void Main(string[] args)
    {
        var c1 = new ConcurrentBag<Guid>();

        Parallel.For(0, 5000000, (i) =>
        {
            c1.Add(NewId.NextGuid());
        });

        var groups = c1.GroupBy(g => g).Where(x => x.Count() != 1).ToList();
        var maxGrouping = groups.Max(x => x.Count());
        
        Console.WriteLine($"Collisions: {groups.Count()}");
        Console.WriteLine($"MaxGroupCollisions: {maxGrouping}");
    }

I know in most cases GUIDs will be generated on different PODs and threads. However, in our case, we have a conversation queue that can only be read by a single service. We can experience anywhere between 450K - 800K messages a daily that need to be logged using unique sequential GUIDs.

Hope this help,

codematrix

from newid.

800k messages a day is nothing, as you said about, you can generate 5 million ids in about 300 milliseconds. And using your exact code, I didn't get a single duplicate in my test, so...

Well, not exact, I used an array (Guid[]) and use result[i] to store the results, instead of adding them to a bag (much slower).

from newid.

There is a develop package that has a ton of tweaks in it if you want to test it out.

from newid.

Sure thing, will do. I'll provide some feedback - it might take a couple days. Just got swamp with a new RFP hence why I didn't respond to the last reply.

from newid.

No worries, I've tweaked performance and moved to a SpinLock, as well as making it possible to ask for a block of ids as an array, to allocate 100's of them at once, versus one call at a time.

from newid.