On Fri, 2023-10-27 at 10:49 +0200, Claudio Jeker wrote: > On Fri, Oct 27, 2023 at 01:54:28AM +0200, Justin Yates Fletcher > wrote: > > On Wed, 2023-10-25 at 20:25 -0400, Raul Miller wrote: > > > On Wed, Oct 25, 2023 at 8:16 PM Justin Yates Fletcher > > > <jyfletc...@gmail.com> wrote: > > > > On Wed, 2023-10-25 at 21:12 +0200, Mike Fischer wrote: > > > > > > > > > > > Am 25.10.2023 um 17:57 schrieb Theo de Raadt > > > > > > <dera...@openbsd.org>: > > > > > > Mike Fischer <fischer+o...@lavielle.com> wrote: > > > > > > > > Am 25.10.2023 um 17:29 schrieb Theo de Raadt > > > > > > We changed a lot of kernel scheduling code *without giving > > > > > > a > > > > > > damn > > > > > > about the stability of this number* > > > > > > > > > > Fine, but you are not changing my running Kernel, are you? > > > > > > > > I don't understand your point with this. Are you making an > > > > accusation? > > > > If not, then why even write this? > > > > > > I think Mike Fischer's point was that the change did not > > > correspond > > > to > > > a kernel upgrade. > > > > > > > It is hyperbole or accusational... or somewhere on that spectrum. > > Either way, it serves no valuable purpose, so why even write that? > > > > Also, there was a kernel change: 7.4. Pretty sure that was > > mentioned. > > > > > > > (And I think Theo de Raadt's point was that there's not enough > > > rigor > > > on load average to diagnose this issue.) > > > > > > > Theo's point, as I read it, was just that the load average is > > calculated in the same way as before, even though there have been > > changes in other parts of the system that could affect it. > > Just to be clear. There was a change in how the load avarage is > calculated. So it may cause differences in numbers. Do we care about > that? > No because it was done to be able to work on more important projects. >
Thanks for the clarification. Maybe I misread. > > > It has nothing to do with rigor. The OS could just always report > > 0.0. > > If you start artifically changing a metric, for the sake of rigor, > > then > > that metric is no longer valuable: > > > > https://en.wikipedia.org/wiki/Goodhart%27s_law > > > > Changing how a mertic is calculated to meet a target certainly > > reduces > > the value of the metric, right? > > I do not agree. The load avarage has some value but most people do > not > understand how it is calculated and what a significant change is. > Also systems change, so metrics change all the time. They still offer > a > good value. > I'm not wanting to get too pedantic, but I'm not quite understanding what you disagree with? My point with this part is just to address the usage of the word "rigor" in regards to calculating the load average. The OP was stating that he would remove load average from his metric collection. I don't think that is a good idea and tried to convey that to him. It is a valuable metric and, like many others, context matters (as you wrote). In response, it was said that Theo implied not enough rigor was applied on load average to diagnose the perceived issue. What I wrote above is in response to specifically that. Goodhart's law popped into my head because it sounded like turning a metric into a target, and the problems of doing so... but maybe I shouldn't have posted that. It might have just confused the point. Anyway, Theo posted a diff on misc@ many years ago (close to 20 maybe?) where the load average would just return 0, in reply to someone complaining about it. So, saying that not enough "rigor" is applied to load average calculation kinda triggered that memory and the response. Justin
