VM's disk doesn't seem to trim

RickyM · Dec 3, 2022

Hi,

My LVM-Thin is getting larger by the day and very fast, I'm almost out of disk space.
I have only one VM (Home Assistant) and nothing else.
Disk space usage only goes up and never goes down.
The discard option is checked and SSD emulation also. I use a SSD, not a HDD.
I found the following guides.

https://gist.github.com/hostberg/86bfaa81e50cc0666f1745e1897c0a56
https://opensource.com/article/20/2/trim-solid-state-storage-linux

But, when I type the command fstrim -av it seems that the VM's drive is not being trimmed.

root@NUC:~# fstrim -av
/boot/efi: 510.7 MiB (535465984 bytes) trimmed on /dev/sda2
/: 231.3 MiB (242548736 bytes) trimmed on /dev/mapper/pve-root

When I look in etc/fstab I don't see the VM's drive, is this why it isn't trimmed?
Should I somehow add the VM's disk here?

# <file system> <mount point> <type> <options> <dump> <pass>
/dev/pve/root / ext4 errors=remount-ro 0 1
UUID=824B-3413 /boot/efi vfat defaults 0 1
/dev/pve/swap none swap sw 0 0
proc /proc proc defaults 0 0

I found the following post, does this refer to etc/fstab? How can I mount the file system within the guest?

https://forum.proxmox.com/threads/trim-ssds.46398/post-220577

LnxBil · Dec 3, 2022

LVM thin uses a much larger extent size in comparision to your filesystem blocksize, so that a simple trim in the guest does not yield a perfectly trimmed result. The internal size is IIRC 2M (512x 4K block) and each extent is "used" if only one single 4K block is in there. So in the worst case, you will not see any trim activity at all. For the best "trimmable" filesystem, I would recommend to use ZFS with a volblocksize that matches your internal filesystem blocksize. With this, you will have a 1:1 mapping and the maximum possible. This will have other drawbacks (and features) but is a completely different beast than LVM-thin.

Besides that, you can also try to overwrite your free space with zeros to implicitly trim the free space. In my experience, this works even better than just trimming in my experience.

RickyM · Dec 3, 2022

Thank you for your suggestions!
I'll look into overwriting the free space with zero and hope this will fix the issue.
The other suggestion seems complicated, I was very happy that I had accomplished this setup

Dunuin · Dec 3, 2022

Did you actually trim inside the guest? "root@NUC:~# fstrim -av" sounds like you try to fstrim on your PVE host and not inside the HomeAssistant VM. Your HomeAssistant OS has to trim/discard too.

RickyM · Dec 3, 2022

I tried the command in HA also, this is the result:

[core-ssh ~]$ fstrim -av
fstrim: unrecognized option: a
BusyBox v1.35.0 (2022-07-18 12:23:02 UTC) multi-call binary.

Usage: fstrim [OPTIONS] MOUNTPOINT

-o OFFSET Offset in bytes to discard from
-l LEN Bytes to discard
-m MIN Minimum extent length
-v Print number of discarded bytes

RickyM · Dec 3, 2022

LnxBil said:
LVM thin uses a much larger extent size in comparision to your filesystem blocksize, so that a simple trim in the guest does not yield a perfectly trimmed result. The internal size is IIRC 2M (512x 4K block) and each extent is "used" if only one single 4K block is in there. So in the worst case, you will not see any trim activity at all. For the best "trimmable" filesystem, I would recommend to use ZFS with a volblocksize that matches your internal filesystem blocksize. With this, you will have a 1:1 mapping and the maximum possible. This will have other drawbacks (and features) but is a completely different beast than LVM-thin.

Besides that, you can also try to overwrite your free space with zeros to implicitly trim the free space. In my experience, this works even better than just trimming in my experience.

It seems that I need to make the disk read only to let zerofree overwrite the free space with zeros, because it can't work with an active file system.
I'm not quite sure how to do this.
Is the last line in fstab correct?

# <file system> <mount point> <type> <options> <dump> <pass>
/dev/pve/root / ext4 errors=remount-ro 0 1
UUID=824B-3413 /boot/efi vfat defaults 0 1
/dev/pve/swap none swap sw 0 0
proc /proc proc defaults 0 0
/dev/sda3 none LVM2_member ro 0 0

I got the type LVM2_member from the command lsblk.

LnxBil · Dec 5, 2022

RickyM said:
It seems that I need to make the disk read only to let zerofree overwrite the free space with zeros, because it can't work with an active file system.
I'm not quite sure how to do this.
Is the last line in fstab correct?

# <file system> <mount point> <type> <options> <dump> <pass>
/dev/pve/root / ext4 errors=remount-ro 0 1
UUID=824B-3413 /boot/efi vfat defaults 0 1
/dev/pve/swap none swap sw 0 0
proc /proc proc defaults 0 0
/dev/sda3 none LVM2_member ro 0 0

I got the type LVM2_member from the command lsblk.

I never use zerofree because of this. Just overwrite the free space inside of your guest with dd:

Code:

dd if=/dev/zero of=/zero bs=64k; sync; sync; sync; rm -f /zero

RickyM · Dec 5, 2022

LnxBil said:
I never use zerofree because of this. Just overwrite the free space inside of your guest with dd:

Code:

dd if=/dev/zero of=/zero bs=64k; sync; sync; sync; rm -f /zero

Thank you!
Noob question: can I just copy/paste this code or do I need to alter it somehow for my own setup?
And just to be sure, when you say "inside of your guest", do you mean inside my NUC's shell or the VM (home assistant) CLI?
I think the NUC is the host and the VM is the guest, just checking to be sure.

LnxBil · Dec 8, 2022

RickyM said:
And just to be sure, when you say "inside of your guest", do you mean inside my NUC's shell or the VM (home assistant) CLI?
I think the NUC is the host and the VM is the guest, just checking to be sure.

Yes, guest = VM (home assistant) in your case.

RickyM said:
can I just copy/paste this code or do I need to alter it somehow for my own setup?

The command assumes a single filesystem layout. You may want to rerun it for every (real) filesystem mounted. Please check with df.

RickyM · Dec 8, 2022

LnxBil said:
Yes, guest = VM (home assistant) in your case.

The command assumes a single filesystem layout. You may want to rerun it for every (real) filesystem mounted. Please check with df.

Thanks!

This is the result of df:

root@NUC:~# df
df: /mnt/pve/PC-Ricky: Host is down
Filesystem            1K-blocks      Used Available Use% Mounted on
udev                    3980044         0   3980044   0% /dev
tmpfs                    802748       996    801752   1% /run
/dev/mapper/pve-root   28465204   3171904  23822020  12% /
tmpfs                   4013728     46800   3966928   2% /dev/shm
tmpfs                      5120         0      5120   0% /run/lock
/dev/sda2                523244       328    522916   1% /boot/efi
/dev/fuse                131072        16    131056   1% /etc/pve
//192.168.1.1/Proxmox 500106236 178475548 321630688  36% /mnt/pve/extbackup
tmpfs                    802744         0    802744   0% /run/user/0

Is this more then one file system? If so, how can I alter your code to rerun for every file system?
I only have one SSD with Proxmox installed and just one VM, Home Assistant.

With df sda3 is not shown, sda3 is the VM, the LVM-Thin drive where I'd like to write the free space with zero's.

root@NUC:~# lsblk
NAME                         MAJ:MIN RM   SIZE RO TYPE MOUNTPOINT
sda                            8:0    0 111.8G  0 disk 
├─sda1                         8:1    0  1007K  0 part 
├─sda2                         8:2    0   512M  0 part /boot/efi
└─sda3                         8:3    0 111.3G  0 part 
  ├─pve-swap                 253:0    0     7G  0 lvm  [SWAP]
  ├─pve-root                 253:1    0  27.8G  0 lvm  /
  ├─pve-data_tmeta           253:2    0     1G  0 lvm  
  │ └─pve-data-tpool         253:4    0  60.7G  0 lvm  
  │   ├─pve-data             253:5    0  60.7G  1 lvm  
  │   ├─pve-vm--100--disk--0 253:6    0     4M  0 lvm  
  │   └─pve-vm--100--disk--1 253:7    0    50G  0 lvm  
  └─pve-data_tdata           253:3    0  60.7G  0 lvm  
    └─pve-data-tpool         253:4    0  60.7G  0 lvm  
      ├─pve-data             253:5    0  60.7G  1 lvm  
      ├─pve-vm--100--disk--0 253:6    0     4M  0 lvm  
      └─pve-vm--100--disk--1 253:7    0    50G  0 lvm  
sdb                            8:16   1  15.3M  0 disk

LnxBil · Dec 8, 2022

RickyM said:
If so, how can I alter your code to rerun for every file system?

This is the HOST not the GUEST

Neobin · Dec 8, 2022

I have no clue of Home Assistant and its underlaying OS, but maybe this helps?:

RickyM · Dec 8, 2022

LnxBil said:
This is the HOST not the GUEST

Ow yes, I'm sorry!

This is the result of df in the guest. Can I just run your code in the guest, or do I need to alter it because there are more file systems?

[core-ssh ~]$ df
Filesystem           1K-blocks      Used Available Use% Mounted on
overlay               50861792  34094940  14663280  70% /
devtmpfs               2005744         0   2005744   0% /dev
tmpfs                  2008048         0   2008048   0% /dev/shm
/dev/sda8             50861792  34094940  14663280  70% /ssl
/dev/sda8             50861792  34094940  14663280  70% /backup
/dev/sda8             50861792  34094940  14663280  70% /share
/dev/sda8             50861792  34094940  14663280  70% /media
/dev/sda8             50861792  34094940  14663280  70% /data
/dev/sda8             50861792  34094940  14663280  70% /config
/dev/sda8             50861792  34094940  14663280  70% /addons
tmpfs                   803220      1500    801720   0% /run/dbus
/dev/sda8             50861792  34094940  14663280  70% /etc/asound.conf
/dev/sda8             50861792  34094940  14663280  70% /run/audio
/dev/sda8             50861792  34094940  14663280  70% /etc/hosts
/dev/sda8             50861792  34094940  14663280  70% /etc/resolv.conf
/dev/sda8             50861792  34094940  14663280  70% /etc/hostname
tmpfs                  2008048         0   2008048   0% /dev/shm
/dev/sda8             50861792  34094940  14663280  70% /etc/pulse/client.conf
tmpfs                  2008048         0   2008048   0% /proc/asound
tmpfs                  2008048         0   2008048   0% /proc/acpi
devtmpfs               2005744         0   2005744   0% /proc/kcore
devtmpfs               2005744         0   2005744   0% /proc/keys
devtmpfs               2005744         0   2005744   0% /proc/timer_list
tmpfs                  2008048         0   2008048   0% /proc/scsi
tmpfs                  2008048         0   2008048   0% /sys/firmware

RickyM · Dec 8, 2022

Neobin said:
I have no clue of Home Assistant and its underlaying OS, but maybe this helps?:

https://github.com/home-assistant/operating-system/issues/994#issuecomment-775039050

https://www.reddit.com/r/Proxmox/comments/y7fc1m/home_assistant_vm_keeps_ballooning_in_size_how

Thank you! I will look in to your information.

RickyM · Dec 21, 2022

Neobin said:
I have no clue of Home Assistant and its underlaying OS, but maybe this helps?:

https://github.com/home-assistant/operating-system/issues/994#issuecomment-775039050

https://www.reddit.com/r/Proxmox/comments/y7fc1m/home_assistant_vm_keeps_ballooning_in_size_how

Just an update:

For me the solution was partly found in the second link.
Only for me the command fstrim -av worked better, but the most important aspect is typing login in the Home Assistant CLI from within Proxmox. A # appears and there I typed fstrim -av.
I also found a folder in Home Assistant called Backup where a lot of backups where compounding, that cleared a lot of space for me also. After deleting those backups I used fstrim again.
To be safe I stopped the core in Home Assistant before running fstrim.

Overwriting the free space with zeros didn't work for some reason. After a while it stops with an error that there is no free space available. No space was freed up after.

Search

Search

VM's disk doesn't seem to trim

RickyM

New Member

LnxBil

Distinguished Member

RickyM

New Member

Dunuin

Distinguished Member

RickyM

New Member

RickyM

New Member

LnxBil

Distinguished Member

RickyM

New Member

LnxBil

Distinguished Member

RickyM

New Member

LnxBil

Distinguished Member

Neobin

Famous Member

RickyM

New Member

RickyM

New Member

RickyM

New Member