AIKIDO-2026-517253

TL;DR

The async read limiter polls pending tasks with a zero timeout and recurses when none have completed. When every task is still pending this becomes a spin loop that consumes a scheduler thread without yielding. Slow async reads on normal request paths can exhaust scheduler resources and degrade availability. The fix waits until at least one task completes instead of busy spinning.