Correctly count real vs. total vs. cores on darwin #672
Conversation
|
👍 ping @chef/client-core |
| cpu[:total] = so.stdout.to_i | ||
| so = shell_out("sysctl -n hw.logicalcpu") | ||
| cpu[:cores] = so.stdout.to_i |
There was a problem hiding this comment.
well, this is confusing. at first glance it seems backwards from what i'd expect.
"cores" should be physical cores -- i would expect '4' here
"total" should be the logical cores things like the scheduler sees - i would expect '8' here
"real" should be how many packages i have plugged into the motherboard - i would expect '1' here
i haven't looked at how we define these on linux or what the APIs return on darwin, but your cores and total read backwards from what i would expect.
(and definitely include some comments if things really need to be confusing).
There was a problem hiding this comment.
checked the darwin APIs and they're the way i assumed they be from the names... so the only remaining question is if our ohai APIs are backwards from what i think they should mean...
|
So on linux the 'total' is the thing that the scheduler sees, so all the hyperthreaded cores aka all the entries in /proc/cpuinfo. So i'm pretty certain 'total' here should match and be |
|
And linux only defines 'real' as the sum of the unique physical_id's which should be the number of sockets. So i'm pretty sure linux behaves the way that I outlined above. |
|
Updated to swap the cores vs. total logic |
|
👍 |
Correctly count real vs. total vs. cores on darwin
We were counting CPUs wrong. See #671 for details