Project

General

Profile

Actions

Support #17271

closed

document that compute nodes need to use cgroupsv1 / cgroupsfs for Docker

Added by Peter Amstutz almost 4 years ago. Updated over 3 years ago.

Status:
Resolved
Priority:
Normal
Assigned To:
Category:
Documentation
Target version:
Due date:
Story points:
-
Release relationship:
Auto

Description

If you have the following versions

linux kernel > 4.x
systemd > 247 (en debian family)
docker > 20.10

Is very likely that you have cgroupsv2 enable by default. You can check by doing

if [ -e /sys/fs/cgroup/cgroup.controllers ] 
then 
  echo "cgroupsV2 enabled" 
fi

Unfortunately Arvados 2.1 doesn't have support for cgrupsV2 so compute nodes need to have this disabled
this can be done by adding `systemd.unified_cgroup_hierarchy=0` to the kernel parameters.

Debian / Ubuntu:

in /etc/default/grub make sure that you have
GRUB_CMDLINE_LINUX="systemd.unified_cgroup_hierarchy=0"

update-grub

Fedora 31

dnf install -y grubby
grubby --update-kernel=ALL --args=”systemd.unified_cgroup_hierarchy=0" 


Subtasks 1 (0 open1 closed)

Task #17280: Review 17271-doc-cgroupv2ResolvedNico César01/26/2021Actions
Actions #1

Updated by Peter Amstutz almost 4 years ago

  • Category set to Documentation
Actions #2

Updated by Peter Amstutz almost 4 years ago

  • Assigned To set to Nico César
Actions #3

Updated by Nico César almost 4 years ago

  • Description updated (diff)
Actions #5

Updated by Nico César almost 4 years ago

  • Status changed from New to In Progress
Actions #6

Updated by Lucas Di Pentima almost 4 years ago

I think it would be useful to explain why cgroups v2 accounting should be turned off, and if the fact of having it enabled makes the system completely fail or just some specific feature to not work properly.
I'm thinking that a preexisting Slurm Cluster admin who is looking for installing Arvados in it may wonder if it's a hard requirement or if they could live with the limitations related to keeping cgroups v2 turned on in the case that it's required by other cluster services.

Other than that, it LGTM.

Actions #7

Updated by Nico César almost 4 years ago

addressing #note-6 4091da044 17271-doc-cgroupv2

Actions #8

Updated by Anonymous almost 4 years ago

  • % Done changed from 0 to 100
  • Status changed from In Progress to Resolved
Actions #9

Updated by Peter Amstutz over 3 years ago

  • Release set to 38
Actions

Also available in: Atom PDF