Incorrect memory reporting

LXD
#1

I remember, while ago, container had shown the right size corresponding with config limits.memory=

It had worked with Ubuntu free -m, with alpine /proc/meminfo.

Now to get back to that status, need to apply lxc config set a1 security.syscalls.intercept.sysinfo=true
to each container and restart them?
sysinfo=true should actually be default, as it was the state of things before lxd version upgrades and it also delivers true information within container, which again speaks for a default situation.

LXD 5.1 has been released
#2

No. Emulation of /proc/meminfo is provided by LXCFS (default enabled in the LXD snap package).

The security.syscalls.intercept.sysinfo is only needed if your application doesn’t rely on /proc/meminfo.

#3

lxd 5.1-4ae3604 23001 latest/stable canonical✓ -

somehow limits.memory=8GiB not working anymore.
Container showing the total server memory, despite:

image.os: ubuntu
image.release: focal
image.serial: “20210108_0856”
limits.memory: 8GiB
security.syscalls.intercept.sysinfo: “true”

after restart:
lxc exec c1 – free -h

              total        used        free      shared  buff/cache   available
 Mem:          125Gi       556Mi       125Gi        85Mi       138Mi       125Gi
 Swap:            0B          0B          0B

host free -h

              total        used        free      shared  buff/cache   available
Mem:          125Gi        62Gi        21Gi       770Mi        42Gi        61Gi
Swap:          29Gi       6.0Mi        29Gi
#4

If you didn’t need security.syscalls.intercept.sysinfo: “true” before to see the memory limits reflected in the container then you should disable it as you won’t need it.

Most likely LXCFS isn’t running, what does ps aux | grep lxcfs on the host show?

#5

root 3122644 0.0 0.0 520552 3572 ? Sl Apr28 1:38 lxcfs /var/snap/lxd/common/var/lib/lxcfs -p /var/snap/lxd/common/lxcfs.pid

#6

Have you tried restarting the container?

#7

I did, I aslo hard restarted lxd (snap restart lxd).

#8

Are you nesting containers?

#9

I’ve not been able to recreate using Focal LXD and container:

root@home02:~# snap install lxd --channel=latest/stable
lxd 5.1-1f6f485 from Canonical✓ installed
root@home02:~# lxd init --auto
root@home02:~# ps aux | grep lxcfs
root        2406  0.0  0.0  85660  1756 ?        Sl   10:05   0:00 lxcfs /var/snap/lxd/common/var/lib/lxcfs -p /var/snap/lxd/common/lxcfs.pid
root        5066  0.0  0.0   6440   720 pts/0    S+   10:05   0:00 grep --color=auto lxcfs
root@home02:~# lxc init images:ubuntu/focal c1
Creating c1
root@home02:~# lxc config set c1 limits.memory=2GiB
root@home02:~# lxc start c1
root@home02:~# lxc exec c1 -- free -m
              total        used        free      shared  buff/cache   available
Mem:           2048          29        2014           0           4        2018
Swap:          2048           0        2048
root@home02:~# free -m
              total        used        free      shared  buff/cache   available
Mem:          11777         366        9926           1        1484       11114
Swap:          4095           0        4095
#10

No, Host is Ubuntu focal 5.4.0-109-generic #123-Ubuntu
unprivileged containers on zfs pool (zfs-0.8.3-1ubuntu12.13.

#11

Does it occur on a freshly created container? It may be something inside the container is unmounting the proc files.

#12

You are right. In freshly created containers, mem is shown right:
lxc exec ub1 – free -h

              total        used        free      shared  buff/cache   available
Mem:          488Mi        40Mi       427Mi       0.0Ki        20Mi       448Mi
Swap:            0B          0B          0B
#13

So most likely something inside the container is unmounting/remounting /proc (possibly systemd)

#14

All existing (old containers) affected, no matter Ubuntu or alpine.
Only the newly fresh created behaved properly.

#15

I’ve moved this (now quite long) discussion to its own topic.