PVC AttachVolume.Attach Failed - Maximum number of block storage volumes are attached
Hey all,
We've been using linode LKE for the past few months and have recently run into this error being reported when running up new pods as soon as any single node hits the maximum number of storage volumes to it:
attachdetach-controller AttachVolume.Attach failed for volume "pvc-a980a27661144deb" : rpc error: code = Internal desc = error attaching volume: [400] Maximum number of block storage volumes are attached to this Linode, please resize Linode to be larger to attach more volumes
Is there any way to force the kube-scheduler to take into account the number of PVC's when it's assigning workloads to nodes instead of assigning it only to find out it can't start afterwards?
Our current workarounds are pretty fragile:
- Delete the pod repeatedly until it moves
- Use nodeSelector to manually assign it to a node
4 Replies
cmcfarland
Linode Staff
Is there any way to force the kube-scheduler to take into account the number of PVC's when it's assigning workloads to nodes instead of assigning it only to find out it can't start afterwards?
There is not a way for this to be done without some larger changes to the kube-scheduler on our side, but I've passed along your specific request to be considered during any discussions around changes or improvements to our LKE service in the future.
I've just run into this same problem as well. Is there any chance it could be addressed as suggested by @BenB ?
Failing that, could the max number of volumes per node be increased?
I have encountered the same issue, is there any solution for the date yet?
internetslang
Linode Staff
Hey there,
As it currently stands, there is still a limit for the amount of volumes that can be attached to a single Node. We are currently still working on increasing this limit, however we do not have an ETA as to when this change will take affect.
For more information on how many volumes can be attached to a single Node, please feel free to read our below documentation:
All the best,
Mike