https://rancher.com/ logo
Join Slack
Powered by
This message was deleted.
# harvester
a
This message was deleted.
b
Is this actually a #CC2UQM49Y question? Afaik this is going to be managed by your storage CSI and not actually harvester, so the answer is going to depend on what you're using to manage/provision storage and not your purely your storageClass.
q
makes sense, i'll ask there too
b
Also, I don't know what you mean by this:
EVERY volume i create must have a volume for it to work properly
Maybe one of those is PVC vs PV ?
q
sorry, every volume i create must have a label.
i think the root of the issue is how L.H. is handling volumes that do not have a label. it seems like if create a volume, and it's not assigned a disk label, it can be placed anywhere. including on disks that do have labels.
b
fwiw There's a lot of re-used terms when it comes to this, but I think this is how it goes: When you add a disk to a VM it' creates a PVC > the storage class creates a volume > The volume creation creates a longhorn volume > The longhorn volume has replicas which it places on disks/nodes depending on settings.
q
that's my impression as well
b
So even though Longhorn mounts a longhorn replica (as part of the longhorn volume) to present it to fulfill the pvc on a node there's actually 4 different object layers it's all filtering through.
q
my understanding is you assign the sc when you create the volume, that sc has a disk label set on it, which matches disks on your server. so then when lh creates the replicas, it would assign it to disks that have the label. but for some reason, replicas w/o a label are also being assigned to the disk.
b
So I think what you're actually asking is "is there a way to prevent Longhorn replicas from being added to a disk with a disk label unless there's a matching label"
👍 1
(I don't know the answer, but just my input based off what you wrote in the other channel)
q
yeah, i appreciate it. framing the question teds to help get the best response.
b
Good luck 🙂
q
thanks!
g
even i am not sure of this, let me check
q
@great-bear-19718 i'm still trying to figure out a solution and have had no luck. docker uses the space available, looks like there's kinda a way to do it with containerd, but i cant figure out how w/ kubernetes / rke2. it's driving me bonkers! 🙂 any help would be great! maybe this will help: https://github.com/containerd/containerd/issues/6694
g
it would help if i understood the issue, your first comment was how to avoid scheduling replicas to unlabelled disks
which is very different from what is the containerd issue
q
oh, i'm sorry, you might have replied to an older post.
that too! lol
that was basically how can i modify a storageclass to have tags basically so i dont have to re-create all of my volumes that do not have tags.
brandond in longhorn chat said: brandondrancher employee you should be able to label the existing disks and add a diskSelector to the storageclass and then put different labels and create a new sc for the high iops disks my understanding is that it’s an opt-in - you can say, require these labels on disks or nodes for this SC. You can’t say, DONT allow these disks to be used by any other SC. (edited)
label everything, use the storageclass to set nodeSelector/diskSelector to put it where you want
containerd issues was: does anyone know how to make a windows container, running on a windows worker node to either see space from mounted volumes, or increase the default size of the root fs? my issue is i created a sqlserver in windows ( i have a specific reason for now to use windows, not linux), but when i restore a database, i keep getting the error that my system only has 20gb of space, dispite restoring to a path that has a pvc mounted to it with 900gb (using local-storage / hostpath)
g
so this is a windows container and not a windows vm?
or just a windows container on a windows vm?
q
windows container on a windows node (harvester vm) inside a harvester down-stream rke2 cluster
g
are you using rke2?
q
yess
e
Hi albert, there is a Longhorn setting which controls whether a volume without any disk selector may have replicas scheduled on a labeled disk: https://longhorn.io/docs/1.7.2/references/settings/#allow-empty-disk-selector-volume As far as I understand, you want to set this to 
false
to disallow unlabeled volumes from having their replicas scheduled on your special disk. This should allow you to keep your existing volumes that don't have disk selectors and also have new volumes with a replica on the special disk, if they carry the corresponding disk selector labels. Whether you use a storage class to make sure the volumes receive the right labels, or you just create volumes directly with the right labels should not make a difference.
q
Thanks @enough-australia-5601 very helpful!
@great-bear-19718 any thoughts on the containerd / windows issue?
3 Views
Open in Slack
PreviousNext
Powered by