One kernel feature, 93% system throughput gone: A Red Hat Enterprise Linux 10.2 kernel regression and how to mitigate it

This blog post discusses a serious kernel regression found in Red Hat Enterprise Linux 10.2 that can lead to a significant drop in system throughput (up to 93%) for large multi-NUMA instances running CockroachDB. The article details the nature of the regression, its impact on performance, and how to mitigate the issue by disabling the NI_RANDOM scheduler feature at runtime. The author presents a thorough investigation of the problem, showcasing personal findings and analysis of various workloads and kernel behaviors.