Channels
academy
amazon
arm
azure
cabpr
chinese
ci-cd
danish
deutsch
developer
elemental
epinio
espanol
events
extensions
fleet
français
gcp
general
harvester
harvester-dev
hobbyfarm
hypper
japanese
k3d
k3os
k3s
k3s-contributor
kim
kubernetes
kubewarden
lima
logging
longhorn-dev
longhorn-storage
masterclass
mesos
mexico
nederlands
neuvector-security
office-hours
one-point-x
onlinemeetup
onlinetraining
opni
os
ozt
phillydotnet
portugues
rancher-desktop
rancher-extensions
rancher-setup
rancher-wrangler
random
rfed_ara
rio
rke
rke2
russian
s3gw
service-mesh
storage
submariner
supermicro-sixsq
swarm
terraform-controller
terraform-provider-rancher2
terraform-provider-rke
theranchcast
training-0110
training-0124
training-0131
training-0207
training-0214
training-1220
ukranian
v16-v21-migration
vsphere
windows
Title
w
wide-easter-7639
05/11/2022, 4:08 PMHi. I am thinking to implement a XCP-ng cluster, and then kuberntes among all nodes. To provide storage to my workloads/pods I am thinking to use longhorn or other solution.
My doubt is if necesary to install longhorn on some nodes dedicating its harddisks specifically to provide storage to other workloads/pods or maybe if it is possible to install longhorn on all the nodes nodes and provide storage to workloads using the local disks.
what is the best practice/approach?
s
sticky-truck-78998
05/12/2022, 6:13 AMI'm no expert but I'll share my two cents. Running it (converged)with workloads on the same nodes hasn't been a problem for us. I think the obvious here is the hardware your trying to use it with when going converged. We only see it use 1-3% on r630’s with dual 14c CPU with 10 SSDs x 3 nodes.
I just caught the hypervisor part. I'm curious for production why hypervisor with a converged approach. Maybe you should check out harvester or kubevirt if you need VMs.
w
wide-easter-7639
05/12/2022, 9:41 AMHi Garet, thanks for your reply.
when you say converged you mean running in the same nodes longhorn, workloads and share the underlying disks via longhorn, right?
👍 1
also I have a doubt. as far as I understand, just to state it: longhorn (or any other storage manager) shares de volumes with the requester pod via network.
I have only used kubernetes over EKS and never had to worried about this.
how do you manage to isolate the workloads network bandwith from the storage network bandwith?
I mean, to avoid the IO throttle the application traffic performance
s
sticky-truck-78998
05/12/2022, 3:20 PMThat feature is hopefully coming in 1.3. We currently run everything on 10G LAGs and haven't had any issues but, this enhancement will follow best practice designs.
Thanks for making me look. Now we have to accommodate the storage network in 1.3 😁 I've been wanting to play with Multis for other reasons like VLAN to Pod.
w
wide-easter-7639
05/13/2022, 5:16 PMGreat to know you're not having issues on a 10G network.
Thanks for your comments!