Thank you, I added "systemd.unified_cgroup_hierarchy=1 " to the kernel's command line and systemd-cgtop now shows Input/Output. However, I ran into a problem with LXC containers after switching to unified mode and though I managed to solve it, I'm worried that something else might break in the future. Is it safe to switch to v2 mode?
On Mon, Dec 5, 2022 at 4:16 PM Michal Koutný <[email protected]> wrote: > Hello. > > On Mon, Dec 05, 2022 at 09:38:18AM +0300, Vladimir Mokrozub < > [email protected]> wrote: > > $ systemctl --version > > systemd 245 (245.4-4ubuntu3.19) > > +PAM +AUDIT +SELINUX +IMA +APPARMOR +SMACK +SYSVINIT +UTMP +LIBCRYPTSETUP > > +GCRYPT +GNUTLS +ACL +XZ +LZ4 +SECCOMP +BLKID +ELFUTILS +KMOD +IDN2 -IDN > > +PCRE2 default-hierarchy=hybrid > ^^^^^^ > Unless you override this on kernel cmdline, it means (blk)io controller > is in v1 mode. > > > systemd-cgtop always has "-" in both Input/s and Output/s columns. There > > are no spikes, even under a high disk load. > > I was testing it with "dd if=/dev/sda of=/dev/null". Here's the output: > > > > Control Group Tasks %CPU Memory Input/s Output/s > > / 214 101.5 3.7G - - > > user.slice 15 99.6 2.9G - - > > system.slice 97 0.4 95.4M - - > > 1) It won't have proper hierarchical behavior (thus no values for .slice > units, cgtop defaults to depth of 3 thus you may not see the active > leaves), > 2) it won't charge writeback IO properly (just FYI, it's not relevant > to your example). > > If you can, I'd suggest you to switch to the unified mode if you want > hierarchical IO accounting. > > HTH, > Michal >
