发起问题

Major-Pickle 的档案 - activity

2020-01-30 00:43:18 -0500 获得奖牌  著名问题 (source)
2020-01-27 06:22:58 -0500 获得奖牌  热门的问题 (source)
2020-01-26 22:59:00 -0500 获得奖牌  受欢迎的问题 (source)
2020-01-26 16:55:21 -0500 问了问题 Trouble with Layer3 network and neutron.

Hi all,

Currently I'm working on a project to setup Openstack Train in a layer 3 routed network. That means the IP the hypervisor (compute node) is assigned is not in the same subnet of where traffic is routed. Further complicating the issue the IP's available to the vms is in another subnet all together.

Each hypervisor has 2 nexthop routes assigned to the subnet. let's say for example the hypervisor has IP 10.2.0.10 in the subnet 10.2.0.0/18. That hypervisor has nexthop routes over the standard interface bridge. eth0 and eth1 which are slaved to br0 each go to separate nexthop routes. We will call these 100.2.5.88 and 100.2.5.89. If we provision an vm it will be assigned an ip from a pool specific to the hypervisor like 10.52.100.0/26. The reason is we have bgp broadcasting the ips as to aid with incoming routing. But we aren't there yet. We can't even get routing off the vm.

No matter what I do with routing inside the vm or on the hypervisor. No matter what settings I pick in neutron I just get nothing. This is with linuxbridge and openvswitch. I've tried letting the routers do it, letting the subnets do it. I've tried neutron segments. I've tried to use the hypervisor as a gateway too. Turning on ipv4 forwarding. But I get nowhere. The network from the vm just doesn't communicate outward.

Simply put letting the hypervisor handle the routing via ipv4 forwarding would be the easiest option. we already have bgp setup too. But at this point I'll take any function. I am using kolla-ansible by the way.

any ideas appreciated. Thank you :)

2015-10-04 00:21:06 -0500 获得奖牌  著名问题 (source)
2015-10-04 00:21:06 -0500 获得奖牌  热门的问题 (source)
2015-10-04 00:21:06 -0500 获得奖牌  受欢迎的问题 (source)
2015-09-22 08:42:24 -0500 问了问题 Provisioning randomly fails

We have a pretty big Openstack infrastructure based on Icehouse. We will upgrade one day. Anyway we have run into a very peculiar problem. 95% of the time we can provision new vm's just fine. We do this via a ruby script by using the Fog module. 5% of the time though the vm fails to provision. It gets stuck in a BUILD state. The VM gets created but the hypervisior and ip are never set. So it seems like the controller knows about it. The compute node that the new VM was intended to go to never gets it. There are no logs at all on that particular compute node. We could then create a new vm and direct it to that compute node to try and rule out problems with it and almost always it will work just fine. In fact we have never had it not work.

We have done so much investigation and rule out all low hanging fruit like not enough resources, return codes, etc. Everything checks out and we are just stuck looking at a blank screen because for all intents and purposes it should have worked.