Closed nurturenature closed 2 years ago
Think it's cgroup v1, v2, hybrid, impedance between host Ubuntu hybrid, docker v2, and container Debain hybrid interactions.
Tried forcing Ubuntu to v2:
# edit /etc/default/grub
# GRUB_CMDLINE_LINUX="systemd.unified_cgroup_hierarchy=1 systemd.legacy_systemd_cgroup_controller=0"
sudo update-grub
sudo reboot
All permutations in docker:
# commented out, :ro, :rw
volumes:
- "/sys/fs/cgroup:/sys/fs/cgroup:ro"
# commented out, true, with & w/o cap_add:
privileged: true
cap_add:
- ALL
With base Ubuntu, privileged:true
hangs and required power cycle :warning:.
With GRUB
explicitly using cgroup v2, container fails with various cgroup issues depending on privileged
, /sys/fs/cgroup
settings.
# look at cgroup usage
grep cgroup /proc/filesystems
cat /proc/self/cgroup
stat -c"%T" -f /sys/fs/cgroup
Tested with current jepsen/master.
Docker logs
sort: cannot read: /var/jepsen/shared/nodes: No such file or directory
Bringing up an individual db node -> :boom: and a power cycle to recover.
LXC is required vs Docker.
Currently have to use
volumes: cgroup
vsprivileged:true
or Ubuntu 20.04 hard hangs when docker-composing.Suspect it's a local dev configuration? (libc6 version or other dependencies that had to be upgraded?)