Hi! 🙂
We detected a blocking call in http-codec module with the help of BlockHound:
This PR fixes the blocking code so the pipeline remains reactive end to end. :) We re-ran the tests and also compared the performance (CPU usage and latency) before and after the fix:
Before
After
Comment From: poutsma
As with #30903, this PR uses the bounded elastic scheduler for all writing operations, including zero-copy operations that do not burden the CPU. I will address this by localizing the usage of the bounded elastic scheduler to just the blocking operations. Doing so will require new method signatures for a public method, so rescheduling to 6.1.