[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Publishing raw generic{,cloud} images without tar, and without compression, plus versionning of point releases



On 2020-05-26 10:29:17 +0200 (+0200), Thomas Goirand wrote:
[...]
> Able to extract the .raw file from a tar.xz? I don't think so. If
> it can, please provide the command line to do that, but as much as
> I know, it's currently not possible. I'd love to be wrong here...
[...]

It *can* be done with Glance's "tasks" API:

https://docs.openstack.org/glance/latest/admin/tasks.html

However that would require every deployment you're using the files
with to have that custom task defined by the Glance administrator.
There are some common tasks which perform image conversion, say
between raw and qcow2, but in general that API was a terrible idea
because any time you expose a Turing-complete interface as an API
you can kiss cross-deployment interoperability guarantees goodbye.

> Now, while what you're saying about RBD may be right, this still
> doesn't help OpenStack users, as the images must be uploaded
> through Glance, and sparse files aren't understood by
> glanceclient.
[...]

Also don't forget that Glance is only part of the story there when
it comes to image transfer and storage. Compute nodes need a copy of
anything they're going to boot (unless you're configured to do
boot-from-volume using copy-on-write with Cinder and Glance sharing
the same backend store), so the larger the logical image size, the
longer it takes to retrieve from Glance or the backend and the more
space it potentially takes in the local image cache. I have seen
image uploads cripple storage networks when lots of compute nodes
try to fetch copies of a new image into their cache all at once. The
bigger the image, the more pronounced the impact from that
thundering herd.
-- 
Jeremy Stanley

Attachment: signature.asc
Description: PGP signature


Reply to: