Bug 2030580

Adding regression as this was fine a few days ago. Three days ago there was a new version, but it only contained a dependency change [1], so apparently something more subtle going on due to the rebuild?

One more data point: `tuned-adm recommend` says "atomic-guest" both when tuned.service is not running (then it additionally warns about "Cannot talk to TuneD daemon via DBus. Is TuneD daemon running?"). In https://github.com/cockpit-project/cockpit/pull/16709 I flip this around to avoid the warning, and was hoping that running tuned would have a better answer -- but it doesn't.

[1] https://gitlab.com/redhat/centos-stream/rpms/tuned/-/commit/23f34274263f7a1364cec0ab1f0b908142ecd8de

This is apparently fixed in 2.17, which is in current RHEL 9.0 nightlies. Thanks!

Sorry, it still happens in current CentOS 9 Stream, and thus also in RHEL 9: http://artifacts.dev.testing-farm.io/c64adb27-6c06-41df-b790-b06c2e497c62/

Still happens in C9s today: https://artifacts.dev.testing-farm.io/2f01d8ca-7599-4d83-adf8-c8082ec31dca/work-allglKM1J/log.txt

It seems it detects it as atomic. The system probably has atomic role in the syspurpose. The TuneD auto-rule is the following:

[atomic-guest]
virt=.+
syspurpose_role=.*atomic.*

I.e. it has to be virtual machine (not bare metal) and it has to have atomic substring in the syspurpose role.

Could you provide output of the following command:
# subscription-manager syspurpose --show

It's just as much effort for me as for you to investigate it on the Testing Farm.. (Good opportunity to fix your c9s tests? [1])

I made a test PR with these commands [2]. In the first run [3] it shows the bug:

+ systemctl status tuned.service
○ tuned.service - Dynamic System Tuning Daemon
     Loaded: loaded (/usr/lib/systemd/system/tuned.service; enabled; vendor preset: enabled)
     Active: inactive (dead)
       Docs: man:tuned(8)
             man:tuned.conf(5)
             man:tuned-adm(8)
+ true
+ systemctl stop tuned.service
+ tuned-adm recommend
Cannot talk to TuneD daemon via DBus. Is TuneD daemon running?
2022-03-18 07:08:22,225 ERROR    tuned.utils.profile_recommender: Failed to process 'syspurpose_role' in '/usr/lib/tuned/recommend.d/50-tuned.conf', the syspurpose module is not available
2022-03-18 07:08:22,260 ERROR    tuned.utils.profile_recommender: Failed to process 'syspurpose_role' in '/usr/lib/tuned/recommend.d/50-tuned.conf', the syspurpose module is not available
atomic-guest
+ subscription-manager syspurpose --show
./run-verify-host.sh: line 9: subscription-manager: command not found
+ true
+ systemctl start tuned.service
+ tuned-adm recommend
atomic-guest
+ subscription-manager syspurpose --show
./run-verify-host.sh: line 13: subscription-manager: command not found
+ true
+ tuned-adm auto_profile
Requested profile 'atomic-guest' doesn't exist.


So it insists on atomic-guest, and subscription-manager is not installed.


[1] https://artifacts.dev.testing-farm.io/6da3a3af-4655-4fc3-86d3-ec1d4561d3de/
[2] https://github.com/cockpit-project/cockpit/pull/17148
[3] https://artifacts.dev.testing-farm.io/ef6aa8fb-55cc-496c-bbaa-d181b6494ffe/

With version 2 [1] I installed subscription-manager, and then the test [2] actually succeeds:


+ systemctl status tuned.service
○ tuned.service - Dynamic System Tuning Daemon
     Loaded: loaded (/usr/lib/systemd/system/tuned.service; enabled; vendor preset: enabled)
     Active: inactive (dead)
       Docs: man:tuned(8)
             man:tuned.conf(5)
             man:tuned-adm(8)
+ true
+ systemctl stop tuned.service
+ tuned-adm recommend
Cannot talk to TuneD daemon via DBus. Is TuneD daemon running?
virtual-guest
+ subscription-manager syspurpose --show
{}
+ systemctl start tuned.service
+ tuned-adm recommend
virtual-guest
+ subscription-manager syspurpose --show
{}
+ tuned-adm auto_profile


So the existence or absence of subscription-manager influences the default profile. sub-man isn't installed on the c9s machines by default, nor on our test (and that makes sense -- it's not useful on centos).

[1] https://github.com/cockpit-project/cockpit/blob/3c620c9314d43ab20a3bee8afccd181f50f4cfc6/test/run-verify-host.sh
[2] https://artifacts.dev.testing-farm.io/35d317e8-e147-410a-8e8b-5193338f9da2/

In version 3 [1] I removed the subscription-manager dependency again, and added `grep -r atomic /etc/ /var/`. The output is in the log [2], but there is nothing *at all* related to subscription-manager or the Atomic product.

This is trivial to reproduce and at that point this requires digging into tuned itself -- I'll leave that to you :-)

Thanks!

[1] https://github.com/cockpit-project/cockpit/blob/fcb4eebac8cf37201459dd3ad08fb4cbb1015b6e/test/run-verify-host.sh
[2] https://artifacts.dev.testing-farm.io/5855d107-fc13-4a73-a4a0-9e3b54f59c58/

This is bug, thanks for reporting, upstream PR:
https://github.com/redhat-performance/tuned/pull/428

I am moving it to the 9.0. It's easier for us to track. The fix should also get to the Cent OS stream.

(In reply to Martin Pitt from comment #1)
> Adding regression as this was fine a few days ago. Three days ago there was
> a new version, but it only contained a dependency change [1], so apparently
> something more subtle going on due to the rebuild?
>
There was no related change on the TuneD side - the last change in the profile recommender was from the 2021-09-22 and in the recommend rules from the 2021-02-09. Before the commit you pointed out it always required subscription-manager on RHEL/CentOS, thus it probably worked because subscription manager was installed and there was no room for the TuneD bug (which was present all the time) to manifest.

> One more data point: `tuned-adm recommend` says "atomic-guest" both when
> tuned.service is not running (then it additionally warns about "Cannot talk
> to TuneD daemon via DBus. Is TuneD daemon running?"). In
> https://github.com/cockpit-project/cockpit/pull/16709 I flip this around to
> avoid the warning, and was hoping that running tuned would have a better
> answer -- but it doesn't.
> 
It should behave the same way no matter whether the TuneD daemon is running or not.

(In reply to Martin Pitt from comment #7)
> It's just as much effort for me as for you to investigate it on the Testing
> Farm.. (Good opportunity to fix your c9s tests? [1])
> 
Our c9s tests should be OK. The test in question failed due to some error on the testing farm which is out of our reach. I.e. in this case it seems the testing farm RPM packages were signed with the wrong GPG key:

Problem opening package beakerlib-1.28-1.20220211112859719800.master.4.g484a443.el9.noarch.rpm. Failing package is: beakerlib-1.28-1.20220211112859719800.master.4.g484a443.el9.noarch
 GPG Keys are configured as: https://download.copr.fedorainfracloud.org/results/packit/beakerlib-beakerlib-master/pubkey.gpg
Error: GPG check FAILED

I.e. the failure had nothing to do with the TuneD or TuneD upstream test suite.

Hello Jaroslav,

> Before the commit you pointed out it always required subscription-manager on RHEL/CentOS

That makes sense -- this would indeed have hidden the bug.

Thanks for fixing this!

For the record: This still breaks in C9S: https://artifacts.dev.testing-farm.io/83437dab-10c1-41af-bd77-73ae1c4d9265/

After evaluating this issue, there are no plans to address it further or fix it in an upcoming release.  Therefore, it is being closed.  If plans change such that this issue will be fixed in an upcoming release, then the bug can be reopened.

Reopening -- this is rather unacceptable. We created a regression in C9S (which is a product!), this has to be fixed. It's bad enough that this has been broken for 8 months already.

Facing similar issue on CentOS 9-Stream baremetal host with or without subscription-manager installed(it detects as atomic-host):-

# rpm -q tuned
tuned-2.18.0-2.el9.noarch

# tuned-adm recommend
atomic-host

# tuned-adm auto_profile
Requested profile 'atomic-host' doesn't exist.

Seems like [1] fixing it as per my local tests, can this be included in tuned rpm?

[1] https://github.com/redhat-performance/tuned/pull/428/commits/1e7dd713346ef1e0b6f6bc602e4a23541681f34a

(In reply to Martin Pitt from comment #18)
> Reopening -- this is rather unacceptable. We created a regression in C9S
> (which is a product!), this has to be fixed. It's bad enough that this has
> been broken for 8 months already.

Sorry, it was closed by automaton.

The fix has been merged upstream some time ago:
https://github.com/redhat-performance/tuned/pull/428

It should be part of the downstream rebase (bug 2057609).

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 (tuned bug fix and enhancement update), 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://access.redhat.com/errata/RHBA-2022:8321