Base VMDK being deployed to incorrect LUN

[colummcgaley]

I'm having a really weird issue.

When a user selects boot from image (creates as a new volume) when launching an instance, the base disk is deployed to an incorrect LUN. Using the option boot from image works fine.

The issue is I never configured Openstack to deploy to SAN2-EUC-VDI-PresistentDisks-LUN21. I don't know where it got the LUN name from, and I don't know how to change it so it no longer does this. The storage components copies the base VMDK to the LUN, and then stores the rest of the VM on the correct LUN, which would be fine if I could control where the base disk is put.

Any ideas would be appreciated.

[squebel]

Same here. Would be nice if a VMware person looking at this forum would answer this question.

[xiangfeiz]

When you use boot from image (creates as a new volume) option, it creates a volume from the image, then boot that volume as an instance. When creating the volume, Cinder selects a datastore considering accessibility, free space etc, and is not restricted to Nova datastore configuration. As a workaround, you can do the following:

1 create a storage policy that points to the Nova datastore(s);

2 create a volume type with vmware:storage_profile equal to the storage policy name;

3 create a volume from image with the volume type;

4 boot the volume to an instance.

[v_bala]

Please mention the OpenStack/VIO version you are using.

There is a way to tell Cinder (using vmware_cluster_name option) to restrict placement to specific compute clusters. But in your case, I think both datastores are connected to hosts in the same compute cluster.

Another option is to use storage policies in volume types as mentioned by xiangfeiz. We can set the cinder default_volume_type to a volume type with the storage policy set. If we do not a specify a volume type, the default volume type will be used. The default_volume_type option is useful for "boot from image (create new volume)" because it may not be possible to specify the volume type from Horizon UI.

FWIW, we also skip datastores in maintenance mode.

[squebel]

v_bala and xiangfeiz, thank you both for your responses. I'll take a deeper look into setting up those storage policies.

We are currently using VIO 2.0 and just into the first parts of a 'POC' for this product. We're really just taking it for a spin and figuring out how it all works and what we might be able to do with it compared to our Mirantis deployment.