Traces disappear after calling gRPC with kotlin coroutine and @WithSpan annotation.

[flexfrank]

Describe the bug

When a suspend function is annotated with @WithSpan and call gRPC, traces disappear after calling gRPC.

Steps to reproduce
Here is an example code.
HelloWorldClient is copied from grpc-kotlin examples.
This logger is configured to display trace and span ids.

import io.grpc.ManagedChannelBuilder
import io.opentelemetry.extension.annotations.WithSpan
import kotlinx.coroutines.runBlocking
import org.slf4j.LoggerFactory

class App {
    companion object {
        private val log = LoggerFactory.getLogger(App::class.java)
    }

    fun run() {
        runBlocking {
            runInner()
        }
    }

    @WithSpan
    private suspend fun runInner() {
        log.info("start run")

        greet()

        log.info("finish run")
    }

    private suspend fun greet() {
        val channel = ManagedChannelBuilder.forAddress("localhost", 50051).usePlaintext().build()
        val client = HelloWorldClient(channel)

        client.use {
            log.info("before gRPC")
            it.greet("world")
            log.info("after gRPC")
        }
    }
}

fun main() {
    App().run()
}

If you annotate run() with @WithSpan instead of runInner(), the logs shows correct traces.

What did you expect to see?
All logs shows the same traces.

What did you see instead?

2022-09-05 15:46:57,329 INFO  [fd4f801b8fdb0d68813b93ec3e6bfe24-9532875b87e3c6f6-01] main App - start run
2022-09-05 15:46:58,044 INFO  [fd4f801b8fdb0d68813b93ec3e6bfe24-9532875b87e3c6f6-01] main App - before gRPC
2022-09-05 15:46:58,822 INFO  [--] main App - after gRPC
2022-09-05 15:46:58,874 INFO  [31410fbdd279d25857f4b0ccaa292ccb-4cd6a61499706ccf-01] main App - finish run

At the "after gRPC" log, the trace disappeared.
The "finish run" log shows newly created trace.

What version are you using?
1.17.0

Environment
Compiler: Corretto-17.0.2.8.1, Kotlin 1.6.20
OS: macOS 12.5.1

Additional context
gRPC Java 1.49.0
gRPC kotlin 1.3.0
protobuf 3.19.4

[andrewloux]

This is the same issue @mateuszrzeszutek #6502

I think @flexfrank's repro is a lot more minimal though; I tried to display a variety of scenarios