"'Using a combination of packet sniffing, #eBPF, and async-profiler we managed to identify the root cause of slow produce requests in our Kafka cluster. We then tested a couple of solutions to the problem: data=writeback journaling mode, fast commits, and changing the file system to XFS.[…] With XFS, the number of produce requests exceeding 65ms (our SLO) was lowered by 82%.'"
