Since 4.12.20, we see that tuned reapplies the entirety of its profiles every few minutes. This also leads to constant restarts of stalld:

~~~
Jun 13 17:08:44 node-name systemd[1]: stalld.service: Succeeded.
Jun 13 17:08:44 node-name systemd[1]: stalld.service: Consumed 247ms CPU time
Jun 13 17:08:44 node-name stalld[1156868]: Disabled RT throttling
Jun 13 17:08:44 node-name stalld[1156870]: lockdown mode is off
Jun 13 17:08:44 node-name stalld[1156870]: /sys/kernel/debug/sched/features doesn't exist
Jun 13 17:08:44 node-name stalld[1156870]: /sys/kernel/debug/sched_features exists
Jun 13 17:08:44 node-name stalld[1156870]: /sys/kernel/debug/sched/debug doesn't exist
Jun 13 17:08:44 node-name stalld[1156870]: /proc/sched_debug exists
Jun 13 17:08:44 node-name stalld[1156870]: boosted pid 0 (undef) using SCHED_DEADLINE
Jun 13 17:08:44 node-name stalld[1156870]: using SCHED_DEADLINE for boosting
Jun 13 17:08:44 node-name stalld[1156870]: initial config_buffer_size set to 4464640
Jun 13 17:08:44 node-name stalld[1156870]: detected new task format
Jun 13 17:08:44 node-name stalld[1156870]: single threaded mode
Jun 13 17:08:49 node-name stalld[1156870]: sched_debug is getting larger, increasing the buffer to 8929280
Jun 13 17:10:19 node-name stalld[1156870]: received signal 15, starting shutdown
Jun 13 17:10:19 node-name stalld[1171052]: Restored RT throttling
Jun 13 17:10:19 node-name systemd[1]: stalld.service: Succeeded.
Jun 13 17:10:19 node-name systemd[1]: stalld.service: Consumed 1.821s CPU time
Jun 13 17:10:19 node-name stalld[1171058]: Disabled RT throttling
Jun 13 17:10:19 node-name stalld[1171060]: lockdown mode is off
Jun 13 17:10:19 node-name stalld[1171060]: /sys/kernel/debug/sched/features doesn't exist
Jun 13 17:10:19 node-name stalld[1171060]: /sys/kernel/debug/sched_features exists
Jun 13 17:10:19 node-name stalld[1171060]: /sys/kernel/debug/sched/debug doesn't exist
Jun 13 17:10:19 node-name stalld[1171060]: /proc/sched_debug exists
Jun 13 17:10:19 node-name stalld[1171060]: boosted pid 0 (undef) using SCHED_DEADLINE
Jun 13 17:10:19 node-name stalld[1171060]: using SCHED_DEADLINE for boosting
Jun 13 17:10:19 node-name stalld[1171060]: initial config_buffer_size set to 4382720
Jun 13 17:10:19 node-name stalld[1171060]: detected new task format
Jun 13 17:10:19 node-name stalld[1171060]: single threaded mode
Jun 13 17:10:24 node-name stalld[1171060]: received signal 15, starting shutdown
Jun 13 17:10:24 node-name stalld[1172133]: Restored RT throttling
~~~

4.12.20

1.
2.
3.

This was already rolled back in 4.14 and 4.13, and it seems that the commit that introduces the issue is not needed and causes more harm than good?
I created and attached a PR, see below