https://rancher.com/ logo
Join Slack
Powered by
This message was deleted.
# harvester
a
This message was deleted.
r
I believe this is currently unachievable as I scanned through KubeVirt’s code base. It doesn’t have any spec definition for network interface queues. However, it’s possible to configure them via KubeVirt’s sidecar mechanism https://github.com/kubevirt/kubevirt/tree/main/cmd/sidecars What @brainy-whale-97450 mentioned is also a good way if you accept the in-band configuration.
t
Thank you, I will look into a sidecar. I responded in the other thread about why ethtool doesn't help in this situation since the max depth is fixed at 256. The setting I reference is how you can increase the max depth of the virtual hardware allowing you to use ethtool inside the VM to bump the depth up.
b
The depth is hardware based. You likely have to issue some command directly to the card via it's configuraiton tool.
t
The virtual hardware has a max depth of 256, that's what the qemu arg is for. The physical has a max depth of 4096 and I already bumped that up to 4096. The physical isn't attached directly to the virtual. physical <-> bridge <-> veth/net* <-> bridge <-> tap <-> virtual
[fedora@fedora-05-02 ~]$ ethtool -g eth1 | grep -v "n/a"
Ring parameters for eth1:
Pre-set maximums:
RX:                     256
TX:                     256
Current hardware settings:
RX:                     256
TX:                     256
TX Push:                off
That's from inside the VM
The simple shell script version of the ConfigMap option on the sidecar page looks promising. I could easily add queue_size options to the driver lines based on their short example.
It was a pain in the neck, but it finally worked. I had to end up building my own sidecar-shim image because KubeVirt 0.54.x included with Harvester doesn't support ConfigMap, so I had to embed my script in a shim image. The stock image didn't seem to include /bin/sh to run a shell script, so I packaged up busybox and enough links to get the script working in my image. This is from inside the VM. The default is set to the max already, I didn't have to run ethtool -G inside the VM to get RX set to 1024. Next week I'll do some more testing to see if that helps my issue any.
Copy code
[root@fedora-05-02 ~]# ethtool -g eth0 | grep -v "n/a"
Ring parameters for eth0:
Pre-set maximums:
RX:                     1024
TX:                     256
Current hardware settings:
RX:                     1024
TX:                     256
TX Push:                off
🙌 2
5 Views
Open in Slack
PreviousNext
Powered by