kubernetes recipe fails to deploy cluster
-
I've probably got something wrong here, but i'm trying out the kubernetes recipe and i'm not having much luck.
Here's what i'm entering in the Recipe dialog
This deploys a master VM, but it takes a while to try and bring up the network then fails to connect and it eventually boots with only a link-local ipv6 (fe80::) address.
This is then followed by loads more cloud-init failures due to lack of network connectivity.
I can't find any docks for this recipe, so i'm kinda guessing. Does this recipe/image require a DHCP server?
-
This is question for @BenjiReis when he's around
-
Does the network CIDR you enter corresponds to the network CIDR of VLAN 15?
If so, you shouldn't do that, Kubernetes master needs its own CIDR to give to its node. -
Thanks @BenjiReis
No, it doesn't - i was using a smaller CIDR.
VLAN 15 is 10.0.15.0/24 and I was trying to get the recipe to assign hosts inside 10.0.15.40/30The Networks in my pools correspond to VLANs defined on my switches.
So i'm not sure how i would proceed, any suggestions? -
Can you try a CIDR not used by any of your networks?
The network would be reserved for the Kubernetes master and nodes.
Here your CIDR overlaps one another. -
@BenjiReis
So using a random unused CIDR results in the same problem.I have to select a Network, the network is either the Default (untagged) or one of the networks that corresponds to a VLAN tag.
Using a CIDR that the switch is not aware of will mean the network won't have a default gateway and no route to the internet, or any of the other VLANs for that matter.
-
No the CIDR used by kubernetes is only used by Kubernetes to communicate between the master and the nodes.
The VMs will get an IP on your network for you to communicate with the VMs. -
@BenjiReis said in kubernetes recipe fails to deploy cluster:
No the CIDR used by kubernetes is only used by Kubernetes to communicate between the master and the nodes.
So each VM gets two VIFs? one private for k8s cluster traffic and another for "public", ingress, node ports etc ?
If thats the case, i'm still not sure this will work for us, as our switches won't be able to route an arbitrary CIDR between XCP hosts if it doesn't belong to one of our networks
The VMs will get an IP on your network for you to communicate with the VMs.
By DHCP ?
-
Perhaps what you can try is having VLAN15 CIDR goes from 10.0.15.0 to 10.0.15.n and the CIDR for kubernetes going from 10.0.15.n+1 to 10.0.15.254 and you're router addressing the whole 10.0.15.X range?
What do you think?
-
@d1rtym0nk3y said in kubernetes recipe fails to deploy cluster:
The VMs will get an IP on your network for you to communicate with the VMs.
By DHCP ?
@BenjiReis could you confirm if the master/nodes get their public address via DHCP ?
-
@d1rtym0nk3y If DHCP is configured on your network then yes.
