Silly test mistake: the throttle was supposed to limit the number of
fragments in the queue, but was closing the throttling ref after
submission rather than transmission so that the queue could grow without
bound sometimes.

It's been like this since the test was introduced in elastic#111933, but it
only started failing recently. I suspect until the performance
improvement in elastic#139988 landed it couldn't actually submit enough tasks
at once to cause a failure.

Closes elastic#143644

Silly test mistake: the throttle was supposed to limit the number of
fragments in the queue, but was closing the throttling ref after
submission rather than transmission so that the queue could grow without
bound sometimes.

It's been like this since the test was introduced in elastic#111933, but it
only started failing recently. I suspect until the performance
improvement in elastic#139988 landed it couldn't actually submit enough tasks
at once to cause a failure.

Closes elastic#143644

Silly test mistake: the throttle was supposed to limit the number of
fragments in the queue, but was closing the throttling ref after
submission rather than transmission so that the queue could grow without
bound sometimes.

It's been like this since the test was introduced in #111933, but it
only started failing recently. I suspect until the performance
improvement in #139988 landed it couldn't actually submit enough tasks
at once to cause a failure.

Closes #143644

Silly test mistake: the throttle was supposed to limit the number of
fragments in the queue, but was closing the throttling ref after
submission rather than transmission so that the queue could grow without
bound sometimes.

It's been like this since the test was introduced in elastic#111933, but it
only started failing recently. I suspect until the performance
improvement in elastic#139988 landed it couldn't actually submit enough tasks
at once to cause a failure.

Closes elastic#143644

Silly test mistake: the throttle was supposed to limit the number of
fragments in the queue, but was closing the throttling ref after
submission rather than transmission so that the queue could grow without
bound sometimes.

It's been like this since the test was introduced in #111933, but it
only started failing recently. I suspect until the performance
improvement in #139988 landed it couldn't actually submit enough tasks
at once to cause a failure.

Closes #143644

Silly test mistake: the throttle was supposed to limit the number of
fragments in the queue, but was closing the throttling ref after
submission rather than transmission so that the queue could grow without
bound sometimes.

It's been like this since the test was introduced in #111933, but it
only started failing recently. I suspect until the performance
improvement in #139988 landed it couldn't actually submit enough tasks
at once to cause a failure.

Closes #143644

Silly test mistake: the throttle was supposed to limit the number of
fragments in the queue, but was closing the throttling ref after
submission rather than transmission so that the queue could grow without
bound sometimes.

It's been like this since the test was introduced in elastic#111933, but it
only started failing recently. I suspect until the performance
improvement in elastic#139988 landed it couldn't actually submit enough tasks
at once to cause a failure.

Closes elastic#143644