Where does raven0 link gets its parentIndex?

[suyuee]

I noticed in our setup the wireguard fails to set up properly sometimes. due to

 error syncing event gw-lambdalabs: fail to ensure wireguar link: error getting default link to: 10.0.95.21, error get vxlan parent with index 0: invalid argument 

So the problems is the parent index is 0;

code pointer:

raven/pkg/networkengine/routedriver/vxlan/utils.go

Lines 261 to 266 in 3197c5d

	if vxLink, ok := link.(*netlink.Vxlan); ok {
	link, err = netlinkutil.LinkByIndex(vxLink.ParentIndex)
	if err != nil {
	return nil, fmt.Errorf("error get vxlan parent: %v", err)
	}
	}

But it seems when creating the vxlan link, the code doesn't actually specify a parentIndex here:

raven/pkg/networkengine/routedriver/vxlan/vxlan.go

Lines 264 to 274 in 3197c5d

	LinkAttrs: netlink.LinkAttrs{
	Name: vxlanLinkName,
	MTU: func(a, b int) int {
	if a > b {
	return b
	}
	return a
	}(vpnDriverMTU, routeDriverMTU),
	Flags: net.FlagUp,
	},
	VxlanId: vxlanID,

When searching the netlink repo, I don't see the parentIndex field set for vxlan links (not sure if it's https://github.com/vishvananda/netlink/blob/77df5d35f725f10c8ca5f5552c95782fd639b46c/link_linux.go#L2098)

Any thoughts? Thanks in advance

[suyuee]

Also seems when there isn't such an error ("error get vxlan parent"...), seems the code path is not triggered and ParentIndex is 0 as well.

[njucjc]

Also seems when there isn't such an error ("error get vxlan parent"...), seems the code path is not triggered and ParentIndex is 0 as well.

Why your default route is point to a vxlan interface？usually it will point to eth0

[suyuee]

ah... I think it's due to our customization, on our edge side, nodes can't talk to each other with their private ip directly; we have to rely on vxlan; that should be why

I think I should just customize it further and do not check parent index of the vxlan? seems it's just never set. Not sure if doing this for raven also makes sense through; the other approach is to set parent index during vxlan creation

[njucjc]

ah... I think it's due to our customization, on our edge side, nodes can't talk to each other with their private ip directly; we have to rely on vxlan; that should be why

I think I should just customize it further and do not check parent index of the vxlan? seems it's just never set. Not sure if doing this for raven also makes sense through; the other approach is to set parent index during vxlan creation

@suyuee I'm not sure if vxlan(raven0) over vxlan (your own vxlan interface) can be supported in raven or not, but the following check can be remove exactly.

 if vxLink, ok := link.(*netlink.Vxlan); ok { 
 	link, err = netlinkutil.LinkByIndex(vxLink.ParentIndex) 
 	if err != nil { 
 		return nil, fmt.Errorf("error get vxlan parent: %v", err) 
 	} 
 } 

can you open a pr to fix it ?

[suyuee]

Thanks for the insights, will open a pr for it. (kinda busy recently, but will do once I get a chance)

Btw, we don't use vxlan over vxlan, but just using raven vxlan fwiw