Ed MacDonald

Submitted [PR 5092](https://github.com/IdentityServer/IdentityServer4/pull/5092)

It's not always a full scan, but it does read many more rows than it needs to (see the query plan linked in the PR). If the optimizer chose to...

@sandcar I see they fixed it in the new code [here](https://github.com/DuendeSoftware/IdentityServer/blob/5f94a2a461d6b8c838c1c14326b1b6b75ff6903b/src/EntityFramework.Storage/TokenCleanup/TokenCleanupService.cs#L91) but just didn't merge this PR for some reason.

First, love this library. We use it for all our batch processes. We have this PR implemented in an internal library to support a few scenarios, but in general it...

I had looked at DataFlow, but channels seemed easier to use and offered everything we needed. Then we discovered this library and things got even easier. I don't think our...

To keep it simple, we limited to two channels. We were thinking of it as the compliment to `Split` but even in a general sense, you could always chain calls:...

> Just for thought, let's consider how .Split should be done. I definitely don't like the out parameter. I'm also not excited about configuring and creating another channel. Those were...

>Experimenting with .Split/.Route... what if one of the channels is closed? If we just consider split (two routes) then I think `Filter` could be the model for half the solution....

Thanks for `Merge` and for thinking about split. If it helps, all we need is a way to get at the "unmatched" items from `Filter`.

Another way to approach this would be to add an overload of `Filter` that accepts a ChannelWriter for unmatched items. The caller would then have to create the channel, but...