[Ci-users] Duffy v2: Please send RFEs!

Thu Aug 24 12:23:21 UTC 2017
Karanbir Singh <mail-lists at karan.org>

On 23/08/17 23:51, Murilo Opsfelder Araújo wrote:
> On 08/22/2017 06:26 AM, Niels de Vos wrote:
>> On Mon, Aug 21, 2017 at 09:55:46PM -0500, Brian Stinson wrote:
>>> Hi Folks,
>>> You're likely familiar with our node provisioner Duffy
>>> (https://wiki.centos.org/QaWiki/CI/Duffy). 
>>> In order to support new features, and encourage interested folks to
>>> participate in development with us, we are gathering requirements to get
>>> started soon on Duffy Version 2. Duffy2 will be developed with an OSS
>>> license, and posted in a public repository.  
>>> Some of you have pending RFEs that we'll deal with during development: 
>>> - Support Fedora Images (https://bugs.centos.org/view.php?id=13626)
>>> - Support for cloud VMs in cloud.ci.centos.org 
>>> - Support for VLAN Isolation on the Seamicro hardware
>>> - Support for extending an existing session
>>>   (https://bugs.centos.org/view.php?id=9773)
>>> If you have a feature request for us to track let's talk about it here.
>> It would be helpful to have an option for keeping extras disks (or a
>> partition) available for free-to-use storage. This can help testing
>> features in Gluster (and probably Ceph) that require a block device. We
>> currently work around that with disk-images/losetup kind of approaches,
>> but it is definitely not how actual deployments are expected to be done.
>> Thanks,
>> Niels
> This feature would also be handy to test partition programs, e.g.,
> blivet and friends.
> Allowing user to specify disk features, for example, block size (512 vs.
> 4096 bytes) and capacity (1G vs. 10G) would also be nice to have.

ofcourse, with a little bit of gymnastics, you can already do this today :)

what would be good to see described, and maybe something for Brian, is
to get the workflow documented for something like this. How I am seeing
this now is that its a case of a custom ks used at deploy time, which in
turn is from a different call ( eg. not a /node/get ), and comes with a
latency question ( ~ 5 min to 10 min ). What needs working out is how
the session management is then done.

maybe something like this :
- session is allocated at the time a call is made ( with no hosts attached )
- machine(s0 are provisioned
- session is populated at the duffy end, with the client ( user /
cico_client etc ) needing to poll for details, say every 60 seconds.

does that sound about right ?

Karanbir Singh
+44-207-0999389 | http://www.karan.org/ | twitter.com/kbsingh
GnuPG Key : http://www.karan.org/publickey.asc