granadol's profile - activity

Sorry. Just read where you don;t want to reduce guest. Please disregard

You change guest MTU by configuring this in the /etc/neutron/dhcp_agent.ini: dnsmasq_config_file=/etc/neutron/dnsmasq-neutron.conf

Then create the dnsmasq-neutron.conf with this:

dhcp-option-force=26,1400

This forces an MTU of 1400

Great. Thank you soooo much for the help.

I want to reduce the ram and vcpu's. Can I create a flavor to do that? I am not certain why my KVM setup will not allow. Nothing special about the setup.

No go. Same error. I cannot understand why.

It does nothing but error with instance not resizing, skipping migration. This happens very quickly and does not register in dashboard.

I have just tried the resize. The hypervisor is KVM. There are errors but nothing that points to a failure. What options should exist in nova.conf to make it happen. No way I have more space because I just resized it up then tried to downsize it.

No errors in ovs_agent logs or openvswitch logs. Current kernel 3.13.0-29-generic

I tried the e1000 to see if it would help the bandwidth issues, defaults is virtio. Speedtest is much faster when running within the qrouter namespace. I cannot get those same results in instances. Tried other images like centos, This instance is ubuntu cloud x64. Anyhelp would be much appreciated.

GRO/TSO are turned off on physical NIC connected to br-ex on network node. Speedtest within the router namspace shows much better bandwidth than instances. No issues are seen using tcpdump/wireshark. Performance is worse when using the e1000 NIC rather than the virtio one.

Can Openstack rezise instances down?

This is listed under the openvswitch plugin however I am using the ml2 plugin. This is where I am confused,

I know when strictly using Openvswitch, promiscuous mode was required on the network adapter being used for br-ex. With Ml2, the documentation does not say. Should the adapter be set with promiscous when using ml2?

I am stumped. I have turned off gro and tso on virtio network adapter in VM. I get on average 10 mbs down and 3 mbps upload. Outside of VM, my upload and download speeds are 49mbps respectively. Intra-vlan traffic is good averaging at 66mbps when performing iperf testing. Internet access to instances are slow. Any ideas or is this normal behavior. I have a dedicated compute, network, controller node. I am using GRE. The host nic's are 1000.

I too have concerns about the network bottleneck. I have a public cloud with openstack and floating public ip's. I have not load tested it with many VM's but it would seem like it could be a problem. I was thinking of possibly implementing LAG on 2 nics to get better throughput and load. Would love to get feedback from a larger implementation.

Sorry it does not sound clear after reading it. I put all commands per the documentation and also added the three commands above the logdir, the state_path and the Lock_path. The others did not need to be added, at least in my setup. If you don't add those three additional commands, you receive errors.

I only put back in these commands and everything works. They are missing in the documentation and cause errors without them.

logdir=/var/log/nova state_path=/var/lib/nova lock_path=/var/lock/nova

The "datastores" tables in database trove needs to be updated with the version ID from the datastore_versions table. Once I did that it started working.

Thank you for this. However, can they be created from within Openstack AKA boot from ISO?

I have seen many conflicting answers to this question. When I try it myself using Cinder, it does not work. Can you create images from ISO in Openstack Icehouse? If so, what is the trick? Thanks.

Juan

I have researched this as well. From what I can find, cloud-archive was used for Ubuntu 12.04. I have seen some references of it being used for Ubuntu 13.10, but not 14.04 LTS. Since 14.04 was built using Openstack Icehouse, I do not believe that the cloud-archive is needed.

Will check on this. Thanks.

It does complain about the 169.254.169.264 being unavailable regardless of instance type Cirros or Ubuntu. Will check. Thank you so much. Beating my head against the wall on this one.

I have used a lot of these troubleshooting steps. I can ping gateways from within the DHCP and router namespaces using the ip netns exec command. I can get to my instance vai ssh when assigned a public ip address. However, from within the instance itself, I cannot ping anything nor get external access. It is almost like the instance did not get added to the namespace. I can see the arp requests for both the instance and gateway. I am very perplexed.

I am having a difficult time troubleshooting my instance issues inability to get to the internet. I can perform an nslookup from the router as well as ping instances. The instances cannot ping the gateways? Should they be able to?