5:04 p.m.
Hello,
I have a local based storage domain that has got full, so that it is now
inactive and virtual machines that are on it are paused (vm paused due to
lack of storage space).
Any advice on how to clean up, eventually deleting some of them?
# df -h /2t_2
Filesystem Size Used Avail Use% Mounted on
/dev/mapper/vg_2t_2-2t_2 1.8T 1.7T 0 100% /2t_2
[root@ovirt01 fdf9546c-68fa-42c9-8a10-78ef3ee534b8]#
Is there any easy way to map disk-to-vm of directories inside
/2t_2/images/dbf9611d-9090-42d6-81e0-58105bc20011/images/ so that I can
"sacrifice" some VMs deleting the corresponding disks' directories, to be
able at least to activate the storage domain again and make a cleaner check?
Or any other suggestions? I'm not able to expand it.
It is not directly managed by me and I suppose too much storage
over-provisioning has been done.
Thanks in advance.
Gianluca
2:55 p.m.
On Fri, May 27, 2022 at 11:04 AM Gianluca Cecchi
<gianluca.cecchi(a)gmail.com> wrote:
Why is the size (1.8T) different from Used (1.7T) despite Use%=100%?
If this is because it's an ext4 filesystem, you can try reducing the
amount reserved to root ('tune2fs -m 1' or so). I think the fs should
be unmounted for this.
I don't believe severe damage can happen by doing this, but really have no idea.
Or any other suggestions? I'm not able to expand it.
Perhaps you can also temporarily move some VMs' disks to some other
local storage and symlink there - no idea what vdsm might do. I think
qemu will be happy, if it can access them (you might need to play with
permissions/selinux/etc).
I'd personally try any of the above on a test setup first.
It is not directly managed by me and I suppose too much storage
over-provisioning has been done.
Good luck and best regards,
--
Didi
3:55 p.m.
On Mon, May 30, 2022 at 7:56 AM Yedidyah Bar David <didi(a)redhat.com> wrote:
Yes, I think the reserved default space.
In the meantime I selected a not important VM where I had a 100Gb data disk
(it was the only one of this type of disk size, so it was easy to find it
on filesystem) and executed
cat /dev/null > disk_file
so that I didn't have any logical inconsistency from oVirt point of view
and 100Gb of space immediately released.
Then I was able to activate the storage domain and delete the VM and go on
with further analysis on disk usage.
Anyway it would be useful to have a way to search through the engine
database to have an easier mapping between disk files and VMs.
Gianluca
11:43 p.m.
New subject: local on host storage domain full, how to clean up
Do you mean something like this ?
/usr/share/ovirt-engine/dbscripts/engine-psql.sh -c "select
disk_id,image_guid,disk_alias,disk_profile_name from all_disks where
disk_content_type='0' order by disk_alias" > /root/disk_ids-to_vms.txt
isi
11:16 p.m.
On Mon, May 30, 2022 at 4:44 PM _isi _ <uli.iske(a)elkb.de> wrote:
partially, because in my case if I take a VM with VM ID
711e5dc6-04a1-42a4-b52c-96685bc19b1f
It has 3 disks, one of which has the alias ostack-ceph1_Disk3
the query for this disk gives
select disk_id,image_guid,disk_alias,disk_profile_name from all_disks
where disk_content_type='0' order by disk_alias;
. . .
1093ada3-65b2-407a-a6b8-fd9529f4a0db |
ded7c7c7-477f-44e3-8ac5-8908bde6340c | ostack-ceph1_Disk3 | 2t_1
And I get under /2t_1
/2t_1/images/caa6fd69-e371-4476-89eb-cbbae82c2add/images
# ll | grep 1093
drwxr-xr-x. 2 vdsm kvm 4096 Apr 20 2021
1093ada3-65b2-407a-a6b8-fd9529f4a0db
#
# cd 1093ada3-65b2-407a-a6b8-fd9529f4a0db/
# ll
total 156872396
-rw-rw----. 1 vdsm kvm 161061273600 Apr 30 2021
ded7c7c7-477f-44e3-8ac5-8908bde6340c
-rw-r--r--. 1 vdsm kvm 289 Apr 20 2021
ded7c7c7-477f-44e3-8ac5-8908bde6340c.meta
#
but I miss the link between the id of the VM and the disk, that is what I
would like to get
1:59 a.m.
New subject: local on host storage domain full, how to clean up
You can use the all_disks_for_vms view, instead of all_disks. Then add
the vm_id and / or vm_names columns.
For example:
/usr/share/ovirt-engine/dbscripts/engine-psql.sh -c "select
disk_id,image_guid,disk_alias,disk_profile_name,vm_id,vm_names from
all_disks_for_vms where disk_content_type='0' order by disk_alias" >
/root/disk_ids-to_vms.txt
-Patrick Hibbs
On Sun, 2022-06-05 at 16:16 +0200, Gianluca Cecchi wrote:
4:34 p.m.
On Sun, Jun 5, 2022 at 7:00 PM Patrick Hibbs <hibbsncc1701(a)gmail.com> wrote:
Thas seems indeed better, thanks!
engine=# select
disk_id,image_guid,disk_alias,disk_profile_name,vm_id,vm_names from
all_disks_for_vms where disk_content_type='0' and
disk_alias='ostack-ceph1_Disk3';
disk_id | image_guid
| disk_alias |
disk_profile_name | vm_id | vm_names
--------------------------------------+--------------------------------------+--------------------+-
------------------+--------------------------------------+--------------
1093ada3-65b2-407a-a6b8-fd9529f4a0db |
ded7c7c7-477f-44e3-8ac5-8908bde6340c | ostack-ceph1_Disk3 |
2t_1 | 711e5dc6-04a1-42a4-b52c-96685bc19b1f | ostack-ceph1
(1 row)
so that I can link the directory and disk file under
/2t_1/images/caa6fd69-e371-4476-89eb-cbbae82c2add/images
1093ada3-65b2-407a-a6b8-fd9529f4a0db/ded7c7c7-477f-44e3-8ac5-8908bde6340c
with the disk alias ostack-ceph1_Disk3
and with the corresponding vm id as seen from GUI
711e5dc6-04a1-42a4-b52c-96685bc19b1f and its name ostack-ceph1
I'm going to crosscheck in a SAN based configuration if any sort
of"translation" is to be done, due to the layout based on LVM and not on
filesystem
Gianluca
1072
days inactive
1083
days old
6 comments
4 participants
participants (4)
-
_isi _ -
Gianluca Cecchi -
Patrick Hibbs -
Yedidyah Bar David