Uploaded image for project: 'RHEL'
  1. RHEL
  2. RHEL-108827

[RHEL-9] Stalld 1.19.8+ doesn't set sched_fifo in systemd unit file

Linking RHIVOS CVEs to...Migration: Automation ...SWIFT: POC ConversionSync from "Extern...XMLWordPrintable

    • Icon: Bug Bug
    • Resolution: Unresolved
    • Icon: Major Major
    • rhel-9.7
    • rhel-9.7
    • stalld
    • stalld-1.20.4-1.el9
    • Yes
    • Important
    • ZStream
    • 1
    • rhel-kernel-rts-time
    • 0
    • False
    • False
    • Hide

      None

      Show
      None
    • Yes
    • CK Parent Issues In Progress
    • Regression Exception
    • Bug Fix
    • Hide
      .Updated the `stalld` scheduling policy regression to prevent performance degradation

      Before this update, the Node Tuning Operator CI was broken because of a change in `stalld` scheduling policy., This change caused the service to revert to SCHED_OTHER instead of SCHED_FIFO after starting. Consequently, real-time workloads could experience performance degradation, and you could not merge PR. With this update, the `systemd` unit file sets `stalld` priority to 10, ensuring that `stalld` runs with SCHED_FIFO. This restores expected behavior and improves performance for real-time workloads.
      Show
      .Updated the `stalld` scheduling policy regression to prevent performance degradation Before this update, the Node Tuning Operator CI was broken because of a change in `stalld` scheduling policy., This change caused the service to revert to SCHED_OTHER instead of SCHED_FIFO after starting. Consequently, real-time workloads could experience performance degradation, and you could not merge PR. With this update, the `systemd` unit file sets `stalld` priority to 10, ensuring that `stalld` runs with SCHED_FIFO. This restores expected behavior and improves performance for real-time workloads.
    • Done
    • Done
    • Done
    • Done
    • All
    • None

      What were you trying to do that didn't work?

      Node Tunining Operator CI is broken.

      What is the impact of this issue to you?

      • customers will likely see performance degradation in real-time workloads
      • cannot merge PRs

      Please provide the package NVR for which the bug is seen:

      sh-5.1# rpm -q stalld
stalld-1.19.8-2.el9_6.x86_64

      How reproducible is this bug?:

      always

      Steps to reproduce

      1. start stalld service
      2. chrt -p $(pidof stalld)

      Expected results

      chrt -p $(pidof stalld)
pid 1151773's current scheduling policy: SCHED_FIFO
pid 1151773's current scheduling priority: 10

      Actual results

      # chrt -p $(pidof stalld)
pid 1151773's current scheduling policy: SCHED_OTHER
pid 1151773's current scheduling priority: 0

       

      Additional information

      We believe that this commit broke stalld.

      We probably want

      ExecStartPost=/usr/bin/chrt -f -p 10 $MAINPID

      added to the systemd unit file.

              wandercosta Wander Costa
              jmencak Jiri Mencak
              Mariya Pershina
              Clark Williams Clark Williams
              Chang Yin Chang Yin
              Krishnaswamy Krishna Kumar Krishnaswamy Krishna Kumar
              Votes:
              0 Vote for this issue
              Watchers:
              11 Start watching this issue

                Created:
                Updated: