On Mo, 29.10.18 13:54, Umut Tezduyar Lindskog (umut@xxxxxxxxxxxx) wrote: > Hey, > > How come the root slice's memory accounting is not matching (or close > to) with what I see in the /sys/fs? Do we do some other special > accounting? > > systemctl 239 (default-hierarchy=hybrid) > > a@b:memory$ pwd > /sys/fs/cgroup/memory > > a@b:memory$ systemctl show -p MemoryCurrent -- -.slice > MemoryCurrent=6363947008 > a@b:memory$ cat memory.usage_in_bytes > 5459550208 > > a@b:memory$ systemctl show -p MemoryCurrent system.slice > MemoryCurrent=1540538368 > a@b:memory$ cat system.slice/memory.usage_in_bytes > 1508548608 In current systemd we'll use the memory accounting exported in /proc/meminfo for the root cgroup. That's because on cgroupsv2 memory is not accounted in the root cgroup at all anymore, because the kernel devs didn't want multiple "sources of truth" anymore. In order to ensure behaviour is somewhat systematic we'll use that data on cgroupsv1 too... The difference in the accounting is probably because one accounts the kernel itself while the other does not. Lennart -- Lennart Poettering, Red Hat _______________________________________________ systemd-devel mailing list systemd-devel@xxxxxxxxxxxxxxxxxxxxx https://lists.freedesktop.org/mailman/listinfo/systemd-devel