PPD-to-TuneD API translation daemon #579

zacikpa · 2023-12-07T14:22:57Z

This PR (currently draft) will introduce a PPD-to-TuneD API translation daemon. Feel free to comment on the implementation during the development process.

tuned/exports/dbus_exporter.py

smallorange · 2023-12-19T07:33:58Z

Hi @zacikpa ,

Thank you for working on this.
I would like to run the patches and test it with GNOME. How could I get it running?

zacikpa · 2023-12-19T07:51:05Z

Hi @smallorange,
this is not yet ready for testing. I'll tag you here when it is.

zacikpa · 2023-12-19T17:22:07Z

@smallorange This is now in a state where it supports profile switching, but it still has some issues.

You can try the following:

Disable PPD:

systemctl disable --now power-profiles-daemon
systemctl mask power-profiles-daemon

In a directory with cloned TuneD, run TuneD manually from the command line:

sudo ./tuned.py

Let TuneD run, open another command line, and run the translation daemon:

sudo ./tuned-ppd.py

So far, I tested this with gnome-control-center. Profile switching works fine, but the daemon does not yet support signalling (and therefore gnome-control-center does not always show the active profile). I haven't yet tested the hold functionality, either.

Once you're done playing with it, don't forget to unmask and enable PPD (if you use it).

smallorange · 2023-12-20T15:31:50Z

Hi @zacikpa,

I've tested it. It works great! I can switch the profile through gnome control center.

Thank you :)

zacikpa · 2024-01-09T15:58:40Z

At this point, the daemon supports all capabilities of PPD except for the "trickle charging" action:

setting and querying the active profile using the ActiveProfile property,
holding and releasing profiles,
querying the current list of holds,
the PerformanceDegraded property, which is set to True when the Intel P-State's no_turbo option is on or when a Thinkpad device is in lap mode. This is currently implemented via trivial polling of sysfs options.

zacikpa · 2024-01-16T12:06:49Z

In the spec file, I now added the daemon as a subpackage of TuneD called tuned-ppd. The daemon exports the same DBus object and interface as PPD, so it conflicts with PPD (though I'm not entirely sure that such conflicts are allowed in Fedora).

This way, applications that currently use PPD can transition to TuneD seamlessly. An alternative option would be to provide a different object (e.g. com.redhat.tuned.PowerProfiles) which would export PPD's interface. That would, however, require changes in those applications.

KyleGospo · 2024-01-17T23:29:47Z

@zacikpa You may also want this:
KyleGospo@f55f1bd

I can say this is working great in my testing!

Conan-Kudo · 2024-01-17T23:34:41Z

tuned.spec

+Requires: %{name} = %{version}
+# The compatibility daemon conflicts with PPD,
+# since it exports the same DBus object.
+Conflicts: power-profiles-daemon


This is fine, but we probably want to adjust both power-profiles-daemon and tuned-ppd packaging to Provide+Conflicts so that it becomes swappable. Maybe something like:

Provides: ppd-service Conflicts: ppd-service

And we'll need to get consumers of power-profiles-daemon to require ppd-service instead.

yarda · 2024-01-18T10:35:32Z

@zacikpa, now when the development of the feature is nearly completed, could you please squash the commits? It would be easier to handle/review and possibly backport.

zacikpa · 2024-01-22T13:12:38Z

@zacikpa You may also want this: KyleGospo@f55f1bd

I can say this is working great in my testing!

Thanks for the patch Kyle, now the daemon can be started as a dependency, e.g., by gnome on startup.

zacikpa · 2024-01-22T13:24:47Z

@zacikpa, now when the development of the feature is nearly completed, could you please squash the commits? It would be easier to handle/review and possibly backport.

I now squashed the changes into 3 commits for review, I can then squash into a single one when approved.

yarda · 2024-01-23T20:15:25Z

@zacikpa, now when the development of the feature is nearly completed, could you please squash the commits? It would be easier to handle/review and possibly backport.

I now squashed the changes into 3 commits for review, I can then squash into a single one when approved.

I think it's OK now, thanks.

yarda

Thanks, good job. It mostly LGTM. I have few minor comments inline and one RFE (configuration file).

I also ran simple tests and it seems it works great.

I noticed it returns high-operating-temperature reason as the degraded perf in case turbo is disabled. In fact turbo can be disabled for various reasons, even manually, i.e. it needn't be due to the high platform temperature. But from the PPD source code, it seems it behaves exactly the same way, i.e. it's probably PPD bug and we should emulate it this (even buggy) way.

tuned/ppd/tuned-ppd.service

tuned-ppd.py

tuned/ppd/controller.py

yarda · 2024-01-24T17:33:47Z

tuned/ppd/controller.py

+LAP_MODE_PATH = "/sys/bus/platform/devices/thinkpad_acpi/dytc_lapmode"
+
+
+class PPDProfile(StrEnum):


Could you use here configuration file with profiles mapping? I think it would be more flexible. The configuration file could be e.g. /etc/tuned/ppd.conf and could e.g. be the INI with e.g. the following format:

[main] base=balanced [hold] min=power-saver max=performance [profiles] # PPD = TuneD power-saver = powersave balanced = balanced performance = throughput-performance

Then the users could customize it or even add more profiles.

Sure, I can do that. What exactly do you propose that the hold section does? Currently, the daemon allows holding only power-saver and performance profiles, similar to PPD.

Sure, I can do that. What exactly do you propose that the hold section does? Currently, the daemon allows holding only power-saver and performance profiles, similar to PPD.

I wanted to map min performance and max performance hold profiles to PPD profile names, but because the mapping to TuneD profiles would be already under the [profiles] section, this is probably redundant abstraction. Let's don't over-engineer it and say power-saver and performance profiles mapping have to be always defined in the [profiles] section.

Otherwise I think it shouldn't be hard change, just one more INI parser and I think it's worth the benefit of greater flexibility.

Right, I agree, let's keep it simple.

tuned/ppd/controller.py

zacikpa · 2024-01-25T11:56:36Z

I think it could shutdown without traceback on the first CTRL+C.

@yarda, I just fixed this using the same method as TuneD does it. Once the profile mapping is moved to a configuration file, I suppose it also makes sense to handle the SIGHUP signal to reload the config.

zacikpa · 2024-01-25T12:32:23Z

Had few reports in our tests about extremely long shutdown times. I was able to reproduce it by reaching the desktop, switching to the performance option in GNOME/KDE, and then shutting down - where it'll hang waiting for dbus to end.

Thanks for the heads up. So far I haven't been able to reproduce on my system, but I'd be happy to investigate once you've got the logs.

yarda · 2024-01-25T12:47:37Z

I think it could shutdown without traceback on the first CTRL+C.

@yarda, I just fixed this using the same method as TuneD does it. Once the profile mapping is moved to a configuration file, I suppose it also makes sense to handle the SIGHUP signal to reload the config.

SIGHUP would be nice :) but it can be also added later.

Do not hardcode Polkit namespace into DBusExporter, provide it as a parameter in the constructor.

zacikpa · 2024-02-06T16:53:18Z

As requested, I changed the implementation to load the profile mapping from a configuration file, which must fulfill three conditions:

the profile mapping must be one-to-one (so that we can translate both ways)
the two PPD profiles that can be held, power-saver and performance, must be present
the default profile must be present in the profile mapping.

zacikpa · 2024-02-06T16:54:57Z

Let me know if I should add any more checks (i.e., warnings when the config file contains unknown options).

yarda

If TuneD is not running it ends with a traceback. I think it could write some legible error message instead:

Traceback (most recent call last):
  File "/usr/lib64/python3.12/site-packages/dbus/bus.py", line 173, in activate_name_owner
    return self.get_name_owner(bus_name)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/lib64/python3.12/site-packages/dbus/bus.py", line 348, in get_name_owner
    return self.call_blocking(BUS_DAEMON_NAME, BUS_DAEMON_PATH,
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/lib64/python3.12/site-packages/dbus/connection.py", line 634, in call_blocking
    reply_message = self.send_message_with_reply_and_block(
                    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
dbus.exceptions.DBusException: org.freedesktop.DBus.Error.NameHasNoOwner: The name does not have an owner

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "/home/yarda/tuned-private/./tuned-ppd.py", line 25, in <module>
    tuned_object = bus.get_object(consts.DBUS_BUS, consts.DBUS_OBJECT)
                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/lib64/python3.12/site-packages/dbus/bus.py", line 237, in get_object
    return self.ProxyObjectClass(self, bus_name, object_path,
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/lib64/python3.12/site-packages/dbus/proxies.py", line 250, in __init__
    self._named_service = conn.activate_name_owner(bus_name)
                          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/lib64/python3.12/site-packages/dbus/bus.py", line 178, in activate_name_owner
    self.start_service_by_name(bus_name)
  File "/usr/lib64/python3.12/site-packages/dbus/bus.py", line 273, in start_service_by_name
    return (True, self.call_blocking(BUS_DAEMON_NAME, BUS_DAEMON_PATH,
                  ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/lib64/python3.12/site-packages/dbus/connection.py", line 634, in call_blocking
    reply_message = self.send_message_with_reply_and_block(
                    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
dbus.exceptions.DBusException: org.freedesktop.DBus.Error.ServiceUnknown: The name is not activatable

Also I don't see the 'ActiveProfile` property in the introspection. E.g. compare introspections of the tuned-ppd and original ppd:

...
+ '    <property type="s" name="ActiveProfile" access="readwrite">\n'
+ '    </property>\n'
...

yarda · 2024-02-07T13:45:00Z

tuned/ppd/daemon.conf

Why not to name it ppd.conf?

yarda · 2024-02-07T13:48:15Z

tuned/ppd/tuned-ppd.conf

This could be just dbus.conf.

yarda · 2024-02-07T14:07:36Z

Let me know if I should add any more checks (i.e., warnings when the config file contains unknown options).

I think the current checks are OK, let's be sane :)

zacikpa · 2024-02-07T15:30:23Z

If TuneD is not running it ends with a traceback. I think it could write some legible error message instead:

Good idea, will fix.

Also I don't see the 'ActiveProfile` property in the introspection. E.g. compare introspections of the tuned-ppd and original ppd.

Yes, python-dbus unfortunately does not support exporting properties via the introspection interface (see discussion in this ancient pull request, TL;DR: the maintainer is against supporting it due to backward compatibility and argues that introspection should only be used for debugging).

tuned-ppd.py

yarda · 2024-02-07T19:48:30Z

Also I don't see the 'ActiveProfile` property in the introspection. E.g. compare introspections of the tuned-ppd and original ppd.

Yes, python-dbus unfortunately does not support exporting properties via the introspection interface (see discussion in this ancient pull request, TL;DR: the maintainer is against supporting it due to backward compatibility and argues that introspection should only be used for debugging).

OK, thanks for the info. Nobody has complained so far, so let's go without it and we will see.

yarda · 2024-02-07T19:54:23Z

LGTM (not counting the unused import found by CodeQL).

The daemon translates DBus API calls to power-profiles-daemon into TuneD API calls.

This includes the configuration of DBus, systemd, Polkit, and a new TuneD Fedora/RHEL subpackage.

zacikpa · 2024-02-07T21:16:12Z

Fixed and squashed.

yarda · 2024-02-07T23:25:21Z

Thanks, merged.

power-profiles-daemon will be replaced by tuned-ppd, see redhat-performance/tuned#579.

and redhat-performance/tuned#579

github-advanced-security bot found potential problems Dec 7, 2023

View reviewed changes

tuned/exports/dbus_exporter.py Fixed Show fixed Hide fixed

ryanabx mentioned this pull request Dec 15, 2023

Idea: Support power-profiles-daemon API linrunner/TLP#720

Closed

zacikpa force-pushed the ppd-daemon branch 5 times, most recently from a6db80b to 359053c Compare December 19, 2023 17:13

zacikpa force-pushed the ppd-daemon branch from 359053c to a46148e Compare December 20, 2023 13:32

zacikpa force-pushed the ppd-daemon branch 2 times, most recently from 7e30a9a to 954d3c1 Compare January 5, 2024 08:50

zacikpa requested a review from yarda January 9, 2024 15:33

KyleGospo mentioned this pull request Jan 12, 2024

feat: Replace power profiles daemon with TuneD ublue-os/bazzite#587

Merged

zacikpa force-pushed the ppd-daemon branch 2 times, most recently from 2d7e2dc to 0ebabbc Compare January 16, 2024 10:34

zacikpa marked this pull request as ready for review January 17, 2024 13:36

Conan-Kudo reviewed Jan 17, 2024

View reviewed changes

zacikpa force-pushed the ppd-daemon branch from 0ebabbc to a58b209 Compare January 22, 2024 12:52

zacikpa force-pushed the ppd-daemon branch from a58b209 to 5579d1a Compare January 22, 2024 13:42

yarda reviewed Jan 24, 2024

View reviewed changes

Make DBusExporter reusable for other services

4502d7a

Do not hardcode Polkit namespace into DBusExporter, provide it as a parameter in the constructor.

zacikpa force-pushed the ppd-daemon branch 4 times, most recently from 7a17c71 to 0ddf1eb Compare February 6, 2024 16:51

zacikpa requested a review from yarda February 6, 2024 16:55

yarda reviewed Feb 7, 2024

View reviewed changes

github-advanced-security bot found potential problems Feb 7, 2024

View reviewed changes

tuned-ppd.py Fixed Show resolved Hide resolved

zacikpa added 2 commits February 7, 2024 22:10

Implement the PPD-to-TuneD daemon

d4b72b2

The daemon translates DBus API calls to power-profiles-daemon into TuneD API calls.

Configure the PPD-to-TuneD daemon to work as a replacement of PPD

3e03b8b

This includes the configuration of DBus, systemd, Polkit, and a new TuneD Fedora/RHEL subpackage.

zacikpa force-pushed the ppd-daemon branch from 34d1405 to 3e03b8b Compare February 7, 2024 21:10

yarda merged commit 11c6c57 into redhat-performance:master Feb 7, 2024
13 of 14 checks passed

Jiri-Koten added a commit to Jiri-Koten/content-resolver-input that referenced this pull request Feb 12, 2024

Remove power-profiles-daemon from RHEL 10

77f4007

power-profiles-daemon will be replaced by tuned-ppd, see redhat-performance/tuned#579.

Jiri-Koten mentioned this pull request Feb 12, 2024

Remove power-profiles-daemon from RHEL 10 minimization/content-resolver-input#1091

Merged

yselkowitz pushed a commit to Jiri-Koten/content-resolver-input that referenced this pull request Feb 12, 2024

Remove power-profiles-daemon from RHEL 10

08a681b

power-profiles-daemon will be replaced by tuned-ppd, see redhat-performance/tuned#579.

yselkowitz pushed a commit to minimization/content-resolver-input that referenced this pull request Feb 12, 2024

Remove power-profiles-daemon from RHEL 10

ee80cb5

power-profiles-daemon will be replaced by tuned-ppd, see redhat-performance/tuned#579.

notroj added a commit to notroj/content-resolver-input that referenced this pull request Feb 13, 2024

Add tuned-ppd replacing power-profiles-daemon, see ee80cb5

f901df3

and redhat-performance/tuned#579

yselkowitz pushed a commit to notroj/content-resolver-input that referenced this pull request Feb 15, 2024

Add tuned-ppd replacing power-profiles-daemon, see ee80cb5

b8af54f

and redhat-performance/tuned#579

yselkowitz pushed a commit to minimization/content-resolver-input that referenced this pull request Feb 15, 2024

Add tuned-ppd replacing power-profiles-daemon, see ee80cb5

9fa6d48

and redhat-performance/tuned#579

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

PPD-to-TuneD API translation daemon #579

PPD-to-TuneD API translation daemon #579

zacikpa commented Dec 7, 2023

smallorange commented Dec 19, 2023

zacikpa commented Dec 19, 2023

zacikpa commented Dec 19, 2023 •

edited

Loading

smallorange commented Dec 20, 2023

zacikpa commented Jan 9, 2024 •

edited

Loading

zacikpa commented Jan 16, 2024

KyleGospo commented Jan 17, 2024

Conan-Kudo Jan 17, 2024

yarda commented Jan 18, 2024

zacikpa commented Jan 22, 2024

zacikpa commented Jan 22, 2024

yarda commented Jan 23, 2024

yarda left a comment •

edited

Loading

yarda Jan 24, 2024

zacikpa Jan 25, 2024

yarda Jan 25, 2024

zacikpa Jan 25, 2024

zacikpa commented Jan 25, 2024 •

edited

Loading

zacikpa commented Jan 25, 2024

yarda commented Jan 25, 2024

zacikpa commented Feb 6, 2024

zacikpa commented Feb 6, 2024

yarda left a comment

yarda Feb 7, 2024

yarda Feb 7, 2024

yarda commented Feb 7, 2024

zacikpa commented Feb 7, 2024

yarda commented Feb 7, 2024

yarda commented Feb 7, 2024

zacikpa commented Feb 7, 2024

yarda commented Feb 7, 2024

		LAP_MODE_PATH = "/sys/bus/platform/devices/thinkpad_acpi/dytc_lapmode"


		class PPDProfile(StrEnum):

PPD-to-TuneD API translation daemon #579

PPD-to-TuneD API translation daemon #579

Conversation

zacikpa commented Dec 7, 2023

smallorange commented Dec 19, 2023

zacikpa commented Dec 19, 2023

zacikpa commented Dec 19, 2023 • edited Loading

smallorange commented Dec 20, 2023

zacikpa commented Jan 9, 2024 • edited Loading

zacikpa commented Jan 16, 2024

KyleGospo commented Jan 17, 2024

Conan-Kudo Jan 17, 2024

Choose a reason for hiding this comment

yarda commented Jan 18, 2024

zacikpa commented Jan 22, 2024

zacikpa commented Jan 22, 2024

yarda commented Jan 23, 2024

yarda left a comment • edited Loading

Choose a reason for hiding this comment

yarda Jan 24, 2024

Choose a reason for hiding this comment

zacikpa Jan 25, 2024

Choose a reason for hiding this comment

yarda Jan 25, 2024

Choose a reason for hiding this comment

zacikpa Jan 25, 2024

Choose a reason for hiding this comment

zacikpa commented Jan 25, 2024 • edited Loading

zacikpa commented Jan 25, 2024

yarda commented Jan 25, 2024

zacikpa commented Feb 6, 2024

zacikpa commented Feb 6, 2024

yarda left a comment

Choose a reason for hiding this comment

yarda Feb 7, 2024

Choose a reason for hiding this comment

yarda Feb 7, 2024

Choose a reason for hiding this comment

yarda commented Feb 7, 2024

zacikpa commented Feb 7, 2024

yarda commented Feb 7, 2024

yarda commented Feb 7, 2024

zacikpa commented Feb 7, 2024

yarda commented Feb 7, 2024

zacikpa commented Dec 19, 2023 •

edited

Loading

zacikpa commented Jan 9, 2024 •

edited

Loading

yarda left a comment •

edited

Loading

zacikpa commented Jan 25, 2024 •

edited

Loading