Hide Forgot
Description of problem: Due to file /lib/udev/tuned-request-ktune-restart sending a USR2 signal to tuned process on device events, settings in <profile_name>/ktune.sh are momentarily reverted to their defaults before being re-applied again. For example, when a VM is started on a RHEV host the LVM logical volume(s) for that VM are activated, which results in performance CPU governor being disabled and then enabled back again. Version-Release number of selected component (if applicable): tuned-0.2.19-16.el6 How reproducible: Always Steps to Reproduce: 1. Start a RHEV VM on that host (or just issue "service ktune condrestart" to simulate the same behaviour). 2. Wait 10 seconds after starting the VM (no need to wait for "service ktune condrestart"). 3. Actual results: In /var/log/tuned/tuned.log : 2016-02-29 11:18:07,434 INFO tuned: performing ktune conditional restart In /var/log/messages : Feb 29 11:18:07 rhevh1-375 cpuspeed: Disabling performance cpu frequency scaling governor Feb 29 11:18:08 rhevh1-375 cpuspeed: Enabling performance cpu frequency scaling governor Expected results: performance cpu governor is not temporarily disabled. Additional info: This is a continuation of bug 1111416, where settings in <profile_name>/sysctl.ktune are preserved on stop / restart. The same is needed for settings in <profile_name>/ktune.sh .
This bug is Ok. Switch to status VERIFIED.
While the proposed patch fixed Tuned and it behaves now correctly, it seems there is still problem with the 'cpuspeed' package which is used by Tuned. I checked cpuspeed code and it seems when it writes: Disabling GOVERNOR cpu frequency scaling governor It reverts governor to the 'userspace' for all CPUs if acpi-cpufreq scaling driver is used or to the 'powersave' if intel_pstate driver is used. Then immediately it sets the governor which is requested (e.g. performance), i.e. the following: Nov 25 05:37:13 hp-dl120gen9-01 cpuspeed: Disabling performance cpu frequency scaling governor Nov 25 05:37:13 hp-dl120gen9-01 cpuspeed: Enabling performance cpu frequency scaling governor Is also bad as it will probably also cause spike. We can either: a) fix cpuspeed b) drop cpuspeed from Tuned Regarding b) Tuned already knows how to set the governor through the sysfs and uses this method as a backup if cpuspeed is not installed.
Problem with cpuspeed was found. Switch this bug to ASSIGNED.
Verified that governor is not switched to another state and then back again. Behaves as expected. Fixed in tuned-0.2.19-18.el6.noarch
Since the problem described in this bug report should be resolved in a recent advisory, it has been closed with a resolution of ERRATA. For information on the advisory, and where to find the updated files, follow the link below. If the solution does not work for you, open a new bug report. https://rhn.redhat.com/errata/RHBA-2017-0738.html