High CPU Usage and Latency reported in Pigz/Gzip commands on VxFS Filesystems with RHEL 9.6

Problem

Multithreaded compression tools such as Pigz and Gzip may exhibit prolonged execution times or become unresponsive on VxFS filesystems running the RHEL 9.6 kernel, often resulting in high CPU utilization

Error Message

There are no specific error message but the command or server could become unresponsive resulting in high CPU utilization.

Cause

The issue relates to changes made in RHEL 9.6 with Non-Uniform Memory Access (NUMA) code

The removal of FOLL_HONOR_NUMA_FAULT in kernel get_user_pages() leads to improper handling of PROT_NONE pages in RHEL9.6, which causes an infinite loop in VxFS buffered write code path

Solution

A supported hotfix has been made available for this issue. Please contact Technical Support to obtain this fix. This hotfix has not yet gone through any extensive Q&A testing. Consequently, if you are not adversely affected by this problem and have a satisfactory temporary workaround in place, we recommend that you wait for the public release of this hotfix.

The Product Engineering Team currently plans to address this issue by way of a patch or hotfix to the current version of the software. Please note that we as a company reserve the right to remove any fix from the targeted release if it does not pass quality assurance tests. Our plans are subject to change and any action taken by you based on the above information or your reliance upon the above information is made at your own risk.

As an interim workaround, the kernel tunable numa_balancing can be disabled until the hotfix is applied. We recommend coordinating with the OS vendor to apply this change.

This issue is also addressed in the upcoming InfoScale 8.0.2 Update 7 patch.