PDA

View Full Version : No valid host was found when launching an image



ocaballero
29-Aug-2014, 14:38
Hi all,

I'm been struggling with SUSE Cloud trying to complete my setup. My environment is a test environment (for demo purposes), and is as follows:

- Host server running openSUSE 13.1 and VMware Workstation 10.
- vmnet10 (NAT) and all VLANs configured in the host server. Communications are OK between SUSE Cloud nodes and host and the internet (using 192.168.124.2 as gateway). Using all default crowbar networks, nothing changed (besides the admin network gateway, as stated before).
- SUSE Cloud version: initially 3 but yesterday I repeated all the process using SUSE Cloud 4.
- All nodes installed as expected and running OK (apparently) as per the deployment guide. I used 1 admin node, 1 control node and 1 compute node (KVM). Control node has 2 HD configured (20 GB each), 4 GB RAM, 1 vCPU, and compute node also has 2 HD configured (50 GB each), 4 GB RAM and 3 vCPU.
- Admin node configured with local SMT. All required repos present and mirrored in there.

As mentioned, I started this journey using SUSE Cloud 3. I got to install all the components in the described environment but when I uploaded a SUSE Studio created QCOW2 image and try to launch it, it always gave me an error (no further information), so I didn't know what else to try.

Then SUSE Cloud 4 went live so I tried the whole thing again using this new version. I got exactly the same issue, but this time SUSE Cloud 4 gave some more specifics about the error:

Message
No valid host was found.
Code
500
Details
File "/usr/lib64/python2.6/site-packages/nova/scheduler/filter_scheduler.py", line 108, in schedule_run_instance raise exception.NoValidHost(reason="")
Created
Aug. 28, 2014, 7:50 p.m.

As my compute node was using KVM, after reading a while I thought that as I was trying to launch a virtual machine (instance) on top of an already virtual machine (compute node over VMware Workstation), maybe I should try to use the QEMU instead of KVM as the nova configuration for the compute node... Tried that, same results :(

My SUSE Studio image, by the way, is a SLES JeOS 64 bits 200 MB size out-of-the-box, brand new created image, downloaded in the OpenStack / KVM format (qcow2).

So, as I ran into the same results repeating over and over the installation steps, even using two different versions of SUSE Cloud, as Einstein mentioned, I must be doing something wrong :confused:.

Does anybody know what could be missing in my scenario and installation? Where else could I look for further information about this error to share with you, if needed?

Thank you so much. Any help will be greatly appreciated.

Oscar

aspiers
31-Aug-2014, 12:31
Most likely you are using the wrong flavor and it doesn't have a suitable amount of RAM or disk for the image or compute hosts in question. For example see https://ask.openstack.org/en/question/33001/openstack-icehouse-instance-launch-error-no-valid-host-was-found/

You should check nova-scheduler.log although last time I looked (one year ago) it required changing nova.conf to log at DEBUG level before this yielded the full explanation. Maybe upstream fixed that since then.

This is certainly a usability issue which needs to be fixed in OpenStack, and in fact it's one I've been meaning to find time for myself for ages, but didn't manage yet.

ocaballero
01-Sep-2014, 13:48
Thank you. I'll try with a bigger flavour and see if that makes any difference.

I'm using a JeOS SLES image (200 MB size) and a flavour defined with 1 vCPU, 512MB RAM and no specified minimum for root and ephemeral disks. I will try increasing that a bit, however I'm positive that those resources should be enough for this image to run.

Have a great week.

Regards
Oscar

aspiers
02-Sep-2014, 12:15
OK cool - if you have no luck with that, remember to try my suggestions about the nova-scheduler.log too!

ocaballero
19-Sep-2014, 02:01
I already used some time to test this. I still have the same problem. Using a flavour with bigger space and RAM does not make any difference.

I saw in the nova-scheduler.log a message saying: "Flavour's disk is too small for requested image". My flavour has 20 GB root disk, 20 GB ephemeral and 2 GB swap with 2 GB RAM. My JeOS SLES image is 300 MB in size, so that flavour amounts must be more than enough.

Completely stuck in here... Any other ideas?

ocaballero
25-Sep-2014, 17:30
Guys,

I finally for this sorted out. As all my environment is using VMware Workstation and Virtual Machines for all nodes, I forgot to enable VT extensions for my CPU in VMware configuration for the Compute Node. Enabling that, everything started working as it should!

Hope this helps others.

Happy Clouding!

Oscar