[Openstack] Create instance fails on creating block device - Block Device Mapping is Invalid

Turbo Fredriksson turbo at bayour.com
Thu Jun 23 13:10:05 UTC 2016


On Jun 23, 2016, at 12:26 PM, Eugen Block wrote:

> /etc/cinder/cinder.conf:enabled_backends = rbd    --> that's what I use currently

"rbd"?

> I'm not sure if it would work, it's been a while since I used local storage, but if you just comment the enabled_backend option out and restart cinder services, I believe it would create local volumes.

Shouldn't it be enough just to "disable" those services/backends?

I guess I have to, because just commenting that out didn't help, they still
show as enabled and running.

But even after disabling them, they're still show as "status=disabled,state=up"
with a "cinder service-list".. ?

>> Ok, that's different! I'm not running Glance on my Compute, only on my Control.
> 
> Glance is not supposed to run on a compute node, it runs on a control node.

Ok, good! I thought I missed something fundamental.

> What's the output of "openstack endpoint list | grep glance"?

| 57b10556b7bf47eaa019c603a0f6b34f | europe-london | glance | image | True | public   | http://10.0.4.1:9292
| 8672f6de1673470d93ab6ccee1c1a2bb | europe-london | glance | image | True | internal | http://10.0.4.1:9292
| e45c3e83fe744e7db949cdd89dfe5654 | europe-london | glance | image | True | admin    | http://10.0.4.1:9292

That's the Control node..

>> [waited a little while]
> 
> How long did you wait?

10-15 seconds perhaps. At least less than (half?) a minute..

>  "This section describes how to install and configure the Image service, code-named glance, on the controller node."

It is not obvious from that that that (!! :) should only be done on the
Controller! It just say "do this on the controller". It does not make it
clear that you shouldn't do something on the compute as well.

> "This section describes how to install and configure the Compute service, code-named nova, on the controller node."
> "This section describes how to install and configure the Compute service on a compute node."

Neither of which distinguish the different parts - what if I
have/want a separate compute and control node? It does not
make things obvious!


And that's why I have a problem with HOWTOs! They _assume_ to much.
And a _BAD_ HOWTO (which all of them on Openstack are!) doesn't even
attempt to explain the different options you have, so if you deviate
even the very slightest, you're f**ked!

There's a _HUMONGOS_ difference between a "HOWTO" and "Documentation"!

> Timeout problem? Make sure that nothing blocks the requests (proxy?), what response do you get if you execute
> control1:~ # curl http://<YOUR-CONTROLLER>:9292

I was doing that ON the Control. Worked just fine.

And The Control and Compute is on the same switch.
-- 
Det är när man känner doften av sin egen avföring
som man börjar undra vem man egentligen är.
- Arne Anka





More information about the Openstack mailing list