https://rancher.com/ logo
Join Slack
Powered by
This message was deleted.
# harvester
a
This message was deleted.
b
Sure but you would end up pinning the VM to that node right?
b
If I can get a VM spun up with direct access to disk, I'll do that or anything else.
Is there a doc or procedure that outlines how to do this somewhere?
b
No it's not a supported thing so it's not in the docs.
b
Gotcha. What's the general idea of me going about this? I can likely figure out the details if you give me a quick outline
b
i know of someone using this: https://github.com/metal-stack/csi-driver-lvm
👀 1
But like I mentioned, not supported.
b
Ok. That's a great lead, thank you!
b
If you get this working please do post any findings / observations. We would be very interested to hear your experiences.
✅ 1
r
Did You try creating StorageClass with locality settings set to "bess-effort" and replica count to 1 (https://docs.harvesterhci.io/v1.3/advanced/storageclass#data-locality-settings). This should give You more or less what You need although IO operations would still go through Longhorn.
b
Definitely. I used strict-local instead to guarantee the speed and it's still limited by the network speed, which is odd. Data xfer shouldn't leave the node but it's still saturating the link
b
Was this through harvester? I thought it prevented you using strict-local
b
For the VM, I created a new storage class, set the keys. Also forwarded the native longhorn UI and ensured the volume was in fact repica=1 with strict-local enabled and verified it was indeed on the same node as the VM
yes, it was in harvester
2 Views
Open in Slack
PreviousNext
Powered by