Issue

What were you trying to do that didn't work? Apply a TuneD profile that sets bootloader kernel arguments (e.g. cpu-partitioning, or a profile that includes it) on a RHEL 9.8 rpm-ostree / "RHEL for Edge" host that is booted the traditional ostree/BLS way (not provisioned or managed by bootc), but which has the stock bootc package installed. tuned-adm profile <profile> returns rc=0 and runtime tuning applies, but the kernel arguments are never staged to the bootloader: rpm-ostree kargs is unchanged and /proc/cmdline does not contain the arguments after reboot. Root cause: in tuned/plugins/plugin_bootloader.py (upstream commit 4baaf40, downstream RHEL-170825 Integration ) the plugin probes bootc loader-entries set-options-for-source --help at init and, if it returns 0, dispatches through the bootc path before the rpm-ostree path (they are mutually exclusive, bootc wins). The probe only tests whether the bootc binary supports the subcommand — not whether the system is actually bootc-managed. Since bootc ≥ v1.15.2 (here bootc-1.16.4-1.el9_8) ships that subcommand and is present in the RHEL 9.8 base image, a traditionally-deployed rpm-ostree host is misclassified as bootc and runs bootc loader-entries set-options-for-source --source tuned, which does not update the rpm-ostree/BLS boot entry the host actually boots from. The kargs are silently lost. Suggested fix: gate the bootc branch on the system actually being bootc-managed (e.g. a booted bootc image per bootc status), not merely on the binary supporting the subcommand. What is the impact of this issue to you? All low-latency / real-time kernel-argument tuning is silently broken on RHEL 9.8 rpm-ostree hosts that carry the stock bootc package. tuned-adm reports success, so nothing surfaces except that the tuning never takes effect — dangerous for production low-latency deployments. It also blocks MicroShift low-latency CI across releases and architectures; we are temporarily pinning tuned to 2.27.0 as a workaround. Please provide the package NVR for which the bug is seen: tuned-2.28.0-1.el9fdp (broken). Last known good: tuned-2.27.0-2.1.20270724git0eb28ac3.el9fdp. Environment, identical across the good/bad boundary: bootc-1.16.4-1.el9_8, rpm-ostree-2026.1-6.el9_8, grub2-tools-2.06-126.el9_8, grubby-8.40-69.el9. Reproduces on x86_64 and aarch64. How reproducible is this bug?: Always (100%). Steps to reproduce On a RHEL 9.8 rpm-ostree/edge host that has the bootc package installed but was deployed/booted via ostree (not bootc install): Confirm bootc supports the subcommand (so tuned will pick the bootc path): bootc loader-entries set-options-for-source --help; echo rc=$? → returns rc=0 Apply a profile that sets bootloader kargs: echo "isolated_cores=1" > /etc/tuned/cpu-partitioning-variables.conf tuned-adm profile cpu-partitioning; echo rc=$? → returns rc=0 rpm-ostree kargs → no new kargs staged systemctl reboot After reboot, check the kernel command line: grep -o 'nohz=on' /proc/cmdline || echo MISSING Expected results The kernel arguments are staged (visible in rpm-ostree kargs) and present in /proc/cmdline after reboot — as happens with tuned-2.27.0 on the same host. Actual results tuned-adm returns 0, but no kargs are staged via rpm-ostree and /proc/cmdline lacks the profile's kernel args after reboot. Step 3 prints MISSING.

Release Notes

No. of Comments

Resolution

Unresolved