[SOLVED] local-lvm shows full but is not full

[Chris-The-Tuner]

Hi together,

i run a 3 Node Cluster with 3 identical Servers and use the internal local storage for the VM's and CT's since we don't need HA.
On 1 Node we see that the local-lvm is showing 97% full but looking inside the 2 CT's running on that server that could never be the case, they have less then 10GB each...
We tried to move one CT to another Node, then the GUI shows correct values.
If we move the CT back its also correct but if 1 CT goes near filling the local-lvm and then back down to ~10GB the GUI seems not to register that the usage dropped.

Is there a way to manually force re-reading of current usage ?
I'm unable to find a Solution here.

Thanks in advance
Greetings from Germany
Chris-The-Tuner

 

[wolfgang]

Hi,

the usage is correct you have to trim the storage inside the container.
Thin-LVM is thin allocated.
So if you delete files inside the CT you have to tell the Thin-LVM it has to release the block.

 

[Chris-The-Tuner]

I'm not sure how i'm going to do that.
Is there a easy way to do that ?

 

[wolfgang]

Is there a easy way to do that ?

https://wiki.archlinux.org/index.php/Solid_state_drive#TRIM

 

[Chris-The-Tuner]

Ok i'm not getting it.
I did check the lvm.conf since that seems to be the logical way (at least for me).
issue_discards is set to 1 already, in the Summary View of the CT running it shows everything correct, but in the Summary View of the Node it shows a nearly full drive.
I'm afraid of loosing Data if i run something i'm not intended to do.
I really think this is a noobie question but i'm just unsure what to do here :/

Greetings Chris

 

[wolfgang]

As an example, if you delete a file with 8KB on a FileSystem the OS must report this to the Storage.
Because the storage does not know how the blocks are used and so they will not be erased automatically.
For this purpose, you have the discard mount option what report deletion to the underlying Storage.
Alternatively, you can run fstrim what scan the storage if some blocks are no more used and tell the Storage to delete the not used blocks.

The point is in the past with spinner disks there was no need of release empty blocks.
When SSD come, the blocks must release because the blocks must reset to keep the performance up.
The same counts for thin allocated storages like Thin-LVM, Ceph, qcow2,.....
If you do not tell the underlying storage that the blocks will no more use the images getting bigger and bigger.

 

[Chris-The-Tuner]

Thanks a lot !
It klicked now, i understand.
I have SSHD's in my Servers so basicly this Task is going to be essential.
Running fstrim / on the CT in question solved the Issue.
Thank you very mutch for your help !

Regards
Chris