RFR: 8239559: Cgroups: Incorrect detection logic on some systems
matthias.baesken at sap.com
Tue Feb 25 14:28:07 UTC 2020
Hi Severin, makes sense !
Maybe you could adjust the comment and add a bit of info about this .
I do not need to see a new webrev .
Best regards, Matthias
> What you are seeing here is a hybrid system. My workstation is
> hybrid as well. Legacy and hybrid systems are being detected as cgroup
> v1. A hybrid systems mostly behaves like a cgroup v1 system. All the
> controllers are mounted via cgroup v1. If if was detected as cgroup v2
> (unified) most of the imposed memory/cpu limits wouldn't be accounted
> > But the comment in
> > http://cr.openjdk.java.net/~sgehwolf/webrevs/JDK-
> > says
> > 102 // For cgroups v2 all controllers need to have zero hierarchy id
> > 103 // and /proc/self/mountinfo needs to have at least one cgroup
> > 104 // mounted.
> > Should this comment be adjusted ?
> > On the system above we have no 0 in /proc/cgroups however it seems
> to me the system supports cgroug v2 ?
> The comment is still correct. There is a difference between a true
> cgroup v2 only system and a hybrid system like the one you are seing.
> We should only detect true cgroup v2 systems as such.
> Does that make sense?
More information about the core-libs-dev