Skip to content
Commit f2a87d5f authored by Douglas Raillard's avatar Douglas Raillard
Browse files

lisa._assets.kmodules.lisa: Rework *_se events

BREAKING CHANGE

1. Remove lisa__uclamp_util_se event in favor of lisa__uclamp_cfs_task:

    * The _se event had some rq-level values (uclamp_min, uclamp_max)
      that should never have been logged in an _se level event.

    * _se events are a mix of task-level and taskgroup-level properties,
      within the realm of CFS. This is replaced with _cfs_task and
      _cfs_tg events.

2. Remove lisa__uclamp_util_cfs in favor of lisa__uclamp_rq:

    * Align the name with lisa__uclamp_cfs_task

    * Remove the CFS-centric properties, which are either not real (e.g.
      uclamp_avg, since the clamp is not applied on CFS util_avg but on
      some other value after merging with RT util) or already logged
      somewhere else (e.g. CFS util)

    * Create lisa__uclamp_rq event for the uclamp_min and uclamp_max
      properties. It currently will only fire on CFS rq PELT updates,
      but it will eventually be migrated to a tracepoint that would fire
      when the uclamp_min/max are changed because of e.g. an RT task.

3. Deprecate lisa__sched_pelt_se in favor of lisa__sched_pelt_cfs_task
   (tasks) and lisa__sched_pelt_cfs_tg (taskgroups):

  This reduces the size of the event and makes it easier to extract the
  relevant information for a given experiment. On top of that, this
  reduces the average size of PELT events, which is quite welcome.

4. Deprecate lisa__sched_util_est_se in favor of
   lisa__sched_util_est_cfs_task and lisa__sched_util_est_cfs_tg:

   Same reasons as for lisa__sched_pelt_se
parent 0f99c977
Loading
Loading
Loading
Loading
0% Loading or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment