[openstack-dev] [ironic] iscsi-deploy vs agent-deploy comparison

Lucas Alvares Gomes lucasagomes at gmail.com
Tue Dec 13 12:15:44 UTC 2016

Hi Pavlo,

On Tue, Dec 13, 2016 at 11:52 AM, Pavlo Shchelokovskyy
<pshchelokovskyy at mirantis.com> wrote:
> Hi all,
> recent comments in some of my patches hinted me that I might be missing some
> understanding of ironic workings, so I'd like to ask the following question:
> What features/possibilities the iscsi-deploy currently has that agent-deploy
> has not (for example comparing pxe_ipmitool and agent_ipmitool drivers)?
> The only thing I'm aware of is that agent-deploy requires user images to be
> accessible over clean no-auth HTTP from inside provisioning network (and in
> integrated deployment that means swift as glance backend) while iscsi-deploy
> does not need this (works with arbitrary glance backend / not reachable from
> provisioning network).
> Are there other advantages of iscsi-deploy compared to agent-deploy?

Very few that I can think of: Deploying with ISCSI requires less RAM
because IPA won't need to convert the user image in-band before
copying it to the disk device*, the ironic-conductor will do the
conversion and writing onto the iSCSI device. And, the iSCSI code may
also help as a base to the boot from volume work.

So, thanks for this email. I think it might be a good time to start
discussing whether we should get rid of the iSCSI approach or not
since it has very little to offer overall.

* If memory is a problem, the agent driver could stream a raw image
directly onto the disk as well.


More information about the OpenStack-dev mailing list