Add flannel's host-gw backend option
Allow configuring Flannel with 3 different backends Magnum deploys k8s/swarm over a dedicated neutron private network, possibly using flannel. Flannel's `host-gw` backend gives the best performance in this topopolgy (private layer2): no packet processing overhead, no reduction to MTU, scales to many hosts as well as the alternatives. The performance difference is significant, see bug for performance numbers for the 3 backend options. Note that part of this change involves relaxing the minion IP spoofing rules to allow traffic from all dynamically-allocated flannel subnets. This is morally equivalent to what we were doing previously with encapsulation - only now neutron is able to see the inner IP header directly. This patch repurposes the label "flannel_use_vxlan" when the network driver is flannel. 1. Rename the label flannel_use_vxlan to flannel_backend 2. Redefine the value of this label from "yes/no" to "udp/vxlan/host-gw" For example, to create a bay model with flannel as network driver: --network-driver flannel --labels flannel_backend=host-gw Other backend options are udp and vxlan. Co-Authored-By: Ton Ngo <ton@us.ibm.com> Partial-Bug: #1518605 Closes-Bug: #1516789 Change-Id: I6d2441664ad1baaca14d0e6ff4bcddbe75bee094changes/66/241866/10
parent
473b99710f
commit
14909f0944
Loading…
Reference in New Issue