The committers listed above are authorized under a signed CLA.

• ✅ login: dmytr / name: Dmytro Iaroslavskyi (dc24d4f)

Hi @laurit, what do you think about the following change to this PR: dmytr#2?

It seems to be more correct if we store thread's active context when fiber is resumed, and restore it when fiber is suspended. Instead of just restoring the root context.

Tests pass, and I've tested this with my test harness - everything still seems to work.

Hi @laurit, what do you think about the following change to this PR: dmytr#2?

It seems to be more correct if we store thread's active context when fiber is resumed, and restore it when fiber is suspended. Instead of just restoring the root context.

@dmytr I think it would work. But as the thread executing the fibers probably should not have a context outside of executing the fiber, and when it has, it is probably a context that was inadvertently leaked there, I'd keep the current approach just because it is a bit simpler. Unless you are suspecting that the thread could have a meaningful context outside of executing fibers that must be preserved I wouldn't change it.

Hi @laurit 👋 Seems like I'm a bit late to the party 😅

I think that you are right and in most cases context outside of the fiber wouldn't matter.

However I looked into the details of how zio-http library is implemented. It provides HTTP server based on Netty. It used to create custom ZIO runtime that was using threads spawned by Netty for its event loop group to execute ZIO effects (link). I guess it was working with the current version of ZIO instrumentation because Netty instrumentation was doing the right thing somehow.

zio-http's implementation was changed recently to use ZIO's own threads for execution. But I'm wondering if there are any libraries out there that would do something similar and it would matter if we restore the right context when fiber is suspended/ended 🤔

@dmytr Even with netty threads I would assume that the thread that is used for fiber doesn't have a meaningful scope. If it had a meaningful scope it would be in the middle of some other operation and getting borrowed to run the fiber would be strange.
If zio-http also needs instrumentation then when this instrumentation is added concurrency tests in our http server/client test suites could surface the issue if there is one.