[openstack-dev] [magnum] Magnum template manage use platform VS others as a type?
Adrian Otto
adrian.otto at rackspace.com
Thu Jul 16 21:51:11 UTC 2015
Kevin,
You make a really good point. Reducing required inputs from users in exchange for a little more setup by cloud operators is a well justified tradeoff. I'm pretty sure flavors in Nova can have tag Metadata added without a nova extension, right? Can someone check to be sure?
If we do have a way to tag flavors, then let's default the value (as you said) to use in cases where the flavor is untagged, and make that configurable as a Magnum config directive. We could also log a warning each time the default is used unless the administrator disables the log notices in our config. That way we have a way to direct them to relevant documentation if they start using Magnum without tagging any flavors first.
We should also mention flavor tagging in our various setup guides with references to detailed instructions.
Let's also make sure that flavor and image args to bay_create also have a configurable default in Magnum for when they are omitted by the user.
Adrian
-------- Original message --------
From: "Fox, Kevin M" <Kevin.Fox at pnnl.gov>
Date: 07/16/2015 1:32 PM (GMT-08:00)
To: "OpenStack Development Mailing List (not for usage questions)" <openstack-dev at lists.openstack.org>
Subject: Re: [openstack-dev] [magnum] Magnum template manage use platform VS others as a type?
Good point.
+1 on server_type. it seems reasonable.
As for the need, I'd really rather not have my users have to know how to map flavors to server_types themselves. Its something they will get wrong at times, and we'll get emails about/spend time explaining.
The lack of heat conditionals has been unpleasant. I know its being worked on now, but not there yet.
In talking with the heat developers, their current recommendation has been, put the conditional stuff as provider resources in different environment files, and make the template generic. (ala http://hardysteven.blogspot.com/2013/10/heat-providersenvironments-101-ive.html). You can then switch out one environment for another to switch things somewhat conditionally then. I'm not sure if this is flexible enough to handle the concern you have though.
But, I think the conditional thing is not the real issue. Whether it supported proper conditionals, it would work with environments, or it would work with seperate templates, any way you slice it, you need some way to fetch which of the choices you want to specify. Either by being specified manually by the user, or some stored mapping in a config file, nova flavor metadata, or flavor mapping stored in the magnum db.
So does the user provide that piece of information or does the admin attach it to the flavor some how? I'm all for the admin doing it, since I can do it when I setup the flavors/magnum and never have to worry about it again. Maybe even support a default = 'vm' so that I only have to go in and tag the ironic flavors as such. That means I only have to worry about tagging 1 or 2 flavors by hand, and the users don't have to do anything. A way better user experience for all involved.
Thanks,
Kevin
________________________________
From: Adrian Otto [adrian.otto at rackspace.com]
Sent: Thursday, July 16, 2015 12:35 PM
To: OpenStack Development Mailing List (not for usage questions)
Subject: Re: [openstack-dev] [magnum] Magnum template manage use platform VS others as a type?
To be clear we have two pursuits on this thread:
1) What to rename bay.blatform to.
2) How we might eliminate the attribute, or replace it with something more intuitive
We have a consensus now on how to address #1. My direction to Kannan is to proceed using server_type as the new attribute name. If anyone disagrees, you can let us know now, or submit a subsequent patch to address that concern, and we can vote on it in Gerrit.
On this subject of potentially eliminating, or replacing this attribute with something else, let’s continue to discuss that.
One key issue is that our current HOT file format does not have any facility for conditional logic evaluation, so if the Bay orchestration differs between various server_type values, we need to select the appropriate value based on the way the bay is created. I’m open to hearing suggestions for implementing any needed conditional logic, if we can put it into a better place.
Adrian
On Jul 16, 2015, at 8:54 AM, Fox, Kevin M <Kevin.Fox at pnnl.gov<mailto:Kevin.Fox at pnnl.gov>> wrote:
Wait... so the issue is if you were to just use nova flavor, you don't have enough information to choose a set of templates that may be more optimal for that flavor type (like vm's or bare metal)? Is this a NaaS vs flatdhcp kind of thing? I just took a quick skim of the heat templates and it wasn't really clear why the template needs to know.
If that sort of thing is needed, maybe allow a heat environment or the template set to be tagged onto nova flavors in Magnum by the admin, and then the user can be concerned only with nova flavors? They are use to dealing with them. Sahara and Trove do some similar things I think.
Thanks,
Kevin
________________________________
From: Hongbin Lu [hongbin.lu at huawei.com<mailto:hongbin.lu at huawei.com>]
Sent: Wednesday, July 15, 2015 8:42 PM
To: OpenStack Development Mailing List (not for usage questions)
Subject: Re: [openstack-dev] [magnum] Magnum template manage use platform VS others as a type?
Kai,
Sorry for the confusion. To clarify, I was thinking how to name the field you proposed in baymodel [1]. I prefer to drop it and use the existing field ‘flavor’ to map the Heat template.
[1] https://review.openstack.org/#/c/198984/6
From: Kai Qiang Wu [mailto:wkqwu at cn.ibm.com]
Sent: July-15-15 10:36 PM
To: OpenStack Development Mailing List (not for usage questions)
Subject: Re: [openstack-dev] [magnum] Magnum template manage use platform VS others as a type?
Hi HongBin,
I think flavors introduces more confusion than nova_instance_type or instance_type.
As flavors not have binding with 'vm' or 'baremetal',
Let me summary the initial question:
We have two kinds of templates for kubernetes now,
(as templates in heat not flexible like programming language, if else etc. And separate templates are easy to maintain)
The two kinds of kubernets templates, One for boot VM, another boot Baremetal. 'VM' or Baremetal here is just used for heat template selection.
1> If used flavor, it is nova specific concept: take two as example,
m1.small, or m1.middle.
m1.small < 'VM' m1.middle < 'VM'
Both m1.small and m1.middle can be used in 'VM' environment.
So we should not use m1.small as a template identification. That's why I think flavor not good to be used.
2> @Adrian, we have --flavor-id field for baymodel now, it would picked up by heat-templates, and boot instances with such flavor.
3> Finally, I think instance_type is better. instance_type can be used as heat templates identification parameter.
instance_type = 'vm', it means such templates fit for normal 'VM' heat stack deploy
instance_type = 'baremetal', it means such templates fit for ironic baremetal heat stack deploy.
Thanks!
Best Wishes,
--------------------------------------------------------------------------------
Kai Qiang Wu (吴开强 Kennan）
IBM China System and Technology Lab, Beijing
E-mail: wkqwu at cn.ibm.com<mailto:wkqwu at cn.ibm.com>
Tel: 86-10-82451647
Address: Building 28(Ring Building), ZhongGuanCun Software Park,
No.8 Dong Bei Wang West Road, Haidian District Beijing P.R.China 100193
--------------------------------------------------------------------------------
Follow your heart. You are miracle!
<image001.gif>Hongbin Lu ---07/16/2015 04:44:14 AM---+1 for the idea of using Nova flavor directly. Why we introduced the “platform” field to indicate “v
From: Hongbin Lu <hongbin.lu at huawei.com<mailto:hongbin.lu at huawei.com>>
To: "OpenStack Development Mailing List (not for usage questions)" <openstack-dev at lists.openstack.org<mailto:openstack-dev at lists.openstack.org>>
Date: 07/16/2015 04:44 AM
Subject: Re: [openstack-dev] [magnum] Magnum template manage use platform VS others as a type?
________________________________
+1 for the idea of using Nova flavor directly.
Why we introduced the “platform” field to indicate “vm” or “baremetel” is that magnum need to map a bay to a Heat template (which will be used to provision the bay). Currently, Magnum has three layers of mapping:
• platform: vm or baremetal
• os: atomic, coreos, …
• coe: kubernetes, swarm or mesos
I think we could just replace “platform” with “flavor”, if we can populate a list of flovars for VM and another list of flavors for baremetal (We may need an additional list of flavors for container in the future for the nested container use case). Then, the new three layers would be:
• flavor: baremetal, m1.small, m1.medium, …
• os: atomic, coreos, ...
• coe: kubernetes, swarm or mesos
This approach can avoid introducing a new field in baymodel to indicate what Nova flavor already indicates.
Best regards,
Hongbin
From: Fox, Kevin M [mailto:Kevin.Fox at pnnl.gov]
Sent: July-15-15 12:37 PM
To: OpenStack Development Mailing List (not for usage questions)
Subject: Re: [openstack-dev] [magnum] Magnum template manage use platform VS others as a type?
Maybe somehow I missed the point, but why not just use raw Nova flavors? They already abstract away irconic vs kvm vs hyperv/etc.
Thanks,
Kevin
________________________________
From: Daneyon Hansen (danehans) [danehans at cisco.com<mailto:danehans at cisco.com>]
Sent: Wednesday, July 15, 2015 9:20 AM
To: OpenStack Development Mailing List (not for usage questions)
Subject: Re: [openstack-dev] [magnum] Magnum template manage use platform VS others as a type?
All,
IMO virt_type does not properly describe bare metal deployments. What about using the compute_driver parameter?
compute_driver = None
(StrOpt) Driver to use for controlling virtualization. Options include: libvirt.LibvirtDriver, xenapi.XenAPIDriver, fake.FakeDriver, baremetal.BareMetalDriver, vmwareapi.VMwareVCDriver, hyperv.HyperVDriver
http://docs.openstack.org/kilo/config-reference/content/list-of-compute-config-options.html
http://docs.openstack.org/developer/ironic/deploy/install-guide.html
From: Adrian Otto <adrian.otto at rackspace.com<mailto:adrian.otto at rackspace.com>>
Reply-To: "OpenStack Development Mailing List (not for usage questions)" <openstack-dev at lists.openstack.org<mailto:openstack-dev at lists.openstack.org>>
Date: Tuesday, July 14, 2015 at 7:44 PM
To: "OpenStack Development Mailing List (not for usage questions)" <openstack-dev at lists.openstack.org<mailto:openstack-dev at lists.openstack.org>>
Subject: Re: [openstack-dev] [magnum] Magnum template manage use platform VS others as a type?
One drawback to virt_type if not seen in the context of the acceptable values, is that it should be set to values like libvirt, xen, ironic, etc. That might actually be good. Instead of using the values 'vm' or 'baremetal', we use the name of the nova virt driver, and interpret those to be vm or baremetal types. So if I set the value to 'xen', I know the nova instance type is a vm, and 'ironic' means a baremetal nova instance.
Adrian
-------- Original message --------
From: Hongbin Lu <hongbin.lu at huawei.com<mailto:hongbin.lu at huawei.com>>
Date: 07/14/2015 7:20 PM (GMT-08:00)
To: "OpenStack Development Mailing List (not for usage questions)" <openstack-dev at lists.openstack.org<mailto:openstack-dev at lists.openstack.org>>
Subject: Re: [openstack-dev] [magnum] Magnum template manage use platform VS others as a type?
I am going to propose a third option:
3. virt_type
I have concerns about option 1 and 2, because “instance_type” and flavor was used interchangeably before [1]. If we use “instance_type” to indicate “vm” or “baremetal”, it may cause confusions.
[1] https://blueprints.launchpad.net/nova/+spec/flavor-instance-type-dedup
Best regards,
Hongbin
From: Kai Qiang Wu [mailto:wkqwu at cn.ibm.com]
Sent: July-14-15 9:35 PM
To: openstack-dev at lists.openstack.org<mailto:openstack-dev at lists.openstack.org>
Subject: [openstack-dev] [magnum] Magnum template manage use platform VS others as a type?
Hi Magnum Guys,
I want to raise this question through ML.
In this patch https://review.openstack.org/#/c/200401/
For some old history reason, we use platform to indicate 'vm' or 'baremetal'.
This seems not proper for that, @Adrian proposed nova_instance_type, and someone prefer other names, let me summarize as below:
1. nova_instance_type 2 votes
2. instance_type 2 votes
3. others (1 vote, but not proposed any name)
Let's try to reach the agreement ASAP. I think count the final votes winner as the proper name is the best solution(considering community diversity).
BTW, If you not proposed any better name, just vote to disagree all, I think that vote is not valid and not helpful to solve the issue.
Please help to vote for that name.
Thanks
Best Wishes,
--------------------------------------------------------------------------------
Kai Qiang Wu (吴开强 Kennan）
IBM China System and Technology Lab, Beijing
E-mail: wkqwu at cn.ibm.com<mailto:wkqwu at cn.ibm.com>
Tel: 86-10-82451647
Address: Building 28(Ring Building), ZhongGuanCun Software Park,
No.8 Dong Bei Wang West Road, Haidian District Beijing P.R.China 100193
--------------------------------------------------------------------------------
Follow your heart. You are miracle! __________________________________________________________________________
OpenStack Development Mailing List (not for usage questions)
Unsubscribe: OpenStack-dev-request at lists.openstack.org?subject:unsubscribe<mailto:OpenStack-dev-request at lists.openstack.org?subject:unsubscribe>
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
__________________________________________________________________________
OpenStack Development Mailing List (not for usage questions)
Unsubscribe: OpenStack-dev-request at lists.openstack.org<mailto:OpenStack-dev-request at lists.openstack.org>?subject:unsubscribe
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstack.org/pipermail/openstack-dev/attachments/20150716/969757be/attachment.html>
More information about the OpenStack-dev
mailing list