API: add support for hotplug/deplug

[yarda]

Extends the runtime API by the two methods: instance_remove_device(instance_name, device_name) instance_add_device(instance_name, device_name)

E.g. consider the following TuneD profile: [cpus_perf] type=cpu devices=cpu1, cpu2 governor=performance

[cpus_idle] type=cpu devices=cpu3, cpu4 governor=ondemand

After the following API calls: instance_remove_device("cpus_idle", "cpu4") instance_add_device("cpus_perf", "cpu4")

It will result in cpu1, cpu2, cpu4 having "performance" governor and cpu3 having "ondemand" governor.

Resolves: rhbz#2113925

Signed-off-by: Jaroslav Škarvada [email protected]

[lgtm-com[bot]]

This pull request introduces 3 alerts when merging e131879f0e7ac1b4251b70daf4f43d9174132327 into 420267fa3463ef4a06885847c7ca4cd24c781dc8 - view on LGTM.com

new alerts:

• 3 for Unused local variable

[lgtm-com[bot]]

This pull request introduces 3 alerts when merging 6116483c697b0440883d1ee9a0e01eb0ed2d1e69 into f43df56085501a90c7bbd12d62b3f298a4e99e30 - view on LGTM.com

new alerts:

• 3 for Unused local variable

---

Heads-up: LGTM.com's PR analysis will be disabled on the 5th of December, and LGTM.com will be shut down ⏻ completely on the 16th of December 2022. It looks like GitHub code scanning with CodeQL is already set up for this repo, so no further action is needed :rocket:. For more information, please check out our post on the GitHub blog.

[jmencak]

I did some early testing and it seems to work for the most part. Where it fails, however, is when we include a parent profile that already provides a [cpu] plug-in settings. For example, try to include=throughput-performance. You'll see:

2022-12-13 13:37:04,768 WARNING  tuned.plugins.base: instance cpus_perf: no matching devices available
2022-12-13 13:37:04,768 INFO     tuned.plugins.plugin_cpu: Latency settings from non-first CPU plugin instance 'cpus_perf' will be ignored.
2022-12-13 13:37:04,769 WARNING  tuned.plugins.base: instance cpus_idle: no matching devices available
2022-12-13 13:37:04,769 INFO     tuned.plugins.plugin_cpu: Latency settings from non-first CPU plugin instance 'cpus_idle' will be ignored.

[jmencak]

A couple of generic comments. Would it make sense to:

• support CPU lists such as we already do for isolated_cores, e.g.: devices=2,4-7 instead of explicitly saying devices=cpu2, cpu4, cpu5, cpu6, cpu7?
• when a device is already in a group, automatically remove it and assign it to a new group when you call instance_add_device ?

[jmencak]

Adding another issue for tracking. When using the following spec:

      [cpus_perf]
      type=cpu
      devices=
      governor=performance

      [cpus_idle]
      type=cpu
      devices=cpu0,cpu1,cpu2,cpu3
      governor=powersave

All devices are assigned to the performance group. I'd expect no devices assigned in the performance group.

[yarda]

I simplified the API, I dropped the deplug/hotplug methods which could lead to races and inconsistent internal states which is not trivial to handle and added more robust method 'instance_acquire_devices(devices, instance_name)'. See the edited comment 0 for description.

[jmencak]

Thank you for the changes, @yarda . I didn't check out your code (yet), but did some testing and I'm getting the following:

   struct {
      boolean false
      string "Instance 'cpu' not found"
   }

When using your example above and running

  dbus-send \
    --system \
    --print-reply \
    --dest="com.redhat.tuned" \
    /Tuned \
    com.redhat.tuned.control.instance_acquire_devices \
    string:cpu1 string:cpus_perf

I also get similar results when using socket API from PR470. Is this expected?

[yarda]

Thank you for the changes, @yarda . I didn't check out your code (yet), but did some testing and I'm getting the following:

   struct {
      boolean false
      string "Instance 'cpu' not found"
   }

When using your example above and running

  dbus-send \
    --system \
    --print-reply \
    --dest="com.redhat.tuned" \
    /Tuned \
    com.redhat.tuned.control.instance_acquire_devices \
    string:cpu1 string:cpus_perf

I also get similar results when using socket API from PR470. Is this expected?

It's bug in the error message, it should be: "Instance 'cpus_perf' not found"

I will PR fix. Do you have the 'cpus_perf' plugin instance defined? Could you provide your testing profile?

[yarda]

I tried with the profile from comment 0:

$ dbus-send --system --print-reply --dest="com.redhat.tuned" /Tuned com.redhat.tuned.control.instance_acquire_devices string:cpu1 string:cpus_perf
method return time=1675729198.283921 sender=:1.30 -> destination=:1.32 serial=7 reply_serial=2
   struct {
      boolean true
      string "OK"
   }

Btw cpu1 is already handled by the instance cpus_perf thus the above API call has no effect.

[jmencak]

Do you have the 'cpus_perf' plugin instance defined? Could you provide your testing profile?

I was using a different profile than I intended. Sorry about that. It seems to work as expected now. Will do a bit more testing.