PBS 3.4.1 not playing nice with qemu-agent (incorrect memory usage reported)

[updatelee]

I have PBS running as a VM in Proxmox, PBS is saving to a NFS drive that is NOT apart of the proxmox server in any way.

Im going to assume this is a 3.4.1 issue because qemu-agent worked well with PBS 3.3

qemu-ga -V = 7.2.15

PVE 8.3 & PBS 3.3 = memory reported correctly
PVE 8.4.1 & PBS 3.3 = memory reported correctly
PVE 8.4.1 & PBS 3.4.1 = memory reported incorrectly
PVE 8.3 & PBS 3.4.1 = memory reported incorrectly

if I reboot PBS, it reports correctly for a bit, then reports incorrectly again, 2am on the graph is when it starts its nightly backup, so Im guessing thats the reason




Any idea's ?

 

[fiona]

Hi,
please share the VM configuration qm config <ID>. What kernel is running inside the VM? Do you have ballooning enabled? See: https://pve.proxmox.com/wiki/Dynamic_Memory_Management#Ballooning

 

[updatelee]

Correct, balooning is being used. 2048-8192, its very much like its being released after, which is strange its working fine in my Ubuntu VM's and Windows VM's

root@Proxmox:~# qm config 100
agent: 1,fstrim_cloned_disks=1
balloon: 2048
boot: order=scsi0;ide2;net0
cores: 4
cpu: host
ide2: local:iso/proxmox-backup-server_3.3-1.iso,media=cdrom,size=1234460K
memory: 8192
meta: creation-qemu=9.0.2,ctime=1742496350
name: PBS
net0: virtio=BC:24:11:1A:DA:A0,bridge=vmbr0,firewall=1
numa: 0
onboot: 1
ostype: l26
scsi0: local2-lvm:vm-100-disk-0,iothread=1,size=32G
scsihw: virtio-scsi-single
smbios1: uuid=e5d05be3-5d55-4654-9e64-555f70c488a0
sockets: 1
vmgenid: 8a4f807b-1d09-48af-8183-2f2d6a296697

 

[fiona]

its very much like its being released after

Sorry, could you elaborate a bit what you mean exactly here? After what?

 

[fiona]

Why do you think that the reported value is incorrect? In the screenshot it can be seen that most of the RAM is being used for buffers/caches and only <300MiB are free. Caches can be dropped manually if you really want to, but this will hurt performance.

Are you using ZFS inside the VM? If yes, what does arcstat show while the usage is high?

 

[updatelee]

Sorry, could you elaborate a bit what you mean exactly here? After what?

I worded that badly. Its not displaying that the memory has been free'd

 

[updatelee]

Why do you think that the reported value is incorrect? In the screenshot it can be seen that most of the RAM is being used for buffers/caches and only <300MiB are free. Caches can be dropped manually if you really want to, but this will hurt performance.

Are you using ZFS inside the VM? If yes, what does arcstat show while the usage is high?

Not using zfs anywhere, host or guests.

hmm ... I think I see what you mean. PBS (debian based) is showing

root@pbs:~# free
               total        used        free      shared  buff/cache   available
Mem:         8134676      498524      243024        1024     7658860     7636152
Swap:        4063228         268     4062960

Ubuntu on the other hand is showing

updatelee@Ubuntu-VM ~ [127]> free
               total        used        free      shared  buff/cache   available
Mem:        28212812     2038676    24498892       99032     2212108    26174136
Swap:        4194300           0     4194300

so with balooning if another VM needs memory then PBS will free some its cached memory ? because right now its looking like that VM is needing ALL its memory and doesnt have any to spare.

 

[fiona]

so with balooning if another VM needs memory then PBS will free some its cached memory ? because right now its looking like that VM is needing ALL its memory and doesnt have any to spare.

You can manually test if the VM would be able to free up the memory when needed, see here: https://pve.proxmox.com/wiki/Dynamic_Memory_Management#Ballooning