Bug#976190: systemd: LXC memory limits messed up after some time

[Michael Biebl]

Am 01.12.20 um 14:29 schrieb Paul Slootman:
> Hi,
> 
> On Tue 01 Dec 2020, Michael Biebl wrote:
>>>
>>> root at dns:~# free
>>>                 total        used        free      shared  buff/cache   available
>>> Mem:        1048576       41660     1006784       48124         132     1006784
>>> Swap:             0           0           0
>>>
>>>
>>> After some time, I have noticed that they are no longer correct; in fact
>>> they are wildly wrong:
>>>
>>> root at dns:~# free
>>>                 total        used        free      shared  buff/cache   available
>>> Mem:    9007199254740991      598372 9007199254108535       85652       34084 9007199254108535
>>> Swap:             0        1692       -1692
>>>
>>
>> I suppose, this is from within the container?
> 
> Yes, correct
> 
>> How do you apply/setup the limits?
> 
> This is in the libvirt container definition:
> 
> <domain type='lxc'>
>    <name>dns</name>
>    <uuid>32bd6117-0082-49eb-b622-87e1de716c88</uuid>
>    <memory unit='KiB'>1048576</memory>
>    <currentMemory unit='KiB'>1048576</currentMemory>
>    <memtune>
>      <hard_limit unit='KiB'>1048576</hard_limit>
>      <soft_limit unit='KiB'>1048576</soft_limit>
>      <swap_hard_limit unit='KiB'>1048576</swap_hard_limit>
>    </memtune>
>    <vcpu placement='static'>1</vcpu>

Let's loop in the the libvirt maintainers.
If I understand Paul correctly (and read the journal log correctly), the 
memory limits applied by libvirt are lost once a `systemctl 
daemon-reload` is happening. I'm not really familiar how libvirt 
interacts with lxc/systemd, so your input would be welcome here.

Michael