Re: [PATCH 2/3] sched_ext: Rename scx_bpf_consume() to scx_bpf_dsq_move_to_local()
From: Tejun Heo
Date: Sun Nov 10 2024 - 14:38:17 EST
Hello,
On Sun, Nov 10, 2024 at 04:09:15PM +0000, Johannes Bechberger wrote:
> Hello all,
>
> the changes look really good and the renaming will make it much easier
> to explain what a simple scheduler does, lowering the entry barrier.
Andrea, Changwoo, Johannes, can you guys reply with Acked-by so that I can
record the consensus in the commit?
> > > * Insert @p into the vtime priority queue of the DSQ identified by @dsq_id.
> > > - * Tasks queued into the priority queue are ordered by @vtime and always
> > > - * consumed after the tasks in the FIFO queue. All other aspects are identical
> > > - * to scx_bpf_dsq_insert().
> > > + * Tasks queued into the priority queue are ordered by @vtime. All other aspects
> > > + * are identical to scx_bpf_dsq_insert().
> > >
> > > I suggest keeping this part, "and always consumed after the tasks
> > > in the FIFO queue." Otherwise, IIRC, there is no place to explain
> > > the priority between FIFO and priority DSQs explicitly.
> > >
> > I think we don't allow anymore to use the FIFO queue and the prio queue
> > at the same time. Maybe we should clarify this here and also mention
> > that we can't use scx_bpf_dsq_insert_vtime() with the built-in DSQs.
>
> I would like to second that we clarify the differences between the built-in
> and the non-builtin DSQs.
Will add that.
> Also: Could we mention that the priority queue is stable? If I remember correctly,
> then tasks with the same priority are scheduled in a FIFO manner.
If it is, that's accidental. I don't think we want to make a guarantee about
ordering stability. It's not like that's going to make any meaningful
difference.
Thanks.
--
tejun