Open Stack

Hi Alexander,

Thanks for your 5c.

Some clarification – I don’t want to change the networking classes so much as how they’re invoked. Currently, Instance.deploy() creates and pushes the heat fragments necessary to create a network (if it’s the first instance to be deployed). This means that even in a derived class of Instance, I can’t add heat elements that refer to the server before deploy() is called, because the server won’t exist as far as Heat is concerned.

It seems strange that Instance.deploy is responsible for creating network elements, and in current usage, it feels like it is not necessary for the heat stack to be pushed first to create the network and then create the instance. One shorter term option might be that that NeutronNetwork by default calls stack.push, but when created through Instance.deploy() does not (because it will be called to instantiate the server). I will also see if it’s possible for the instance to ask its applications for their softwareconfig elements before it deploys itself, though I’m not sure yet if I like that usage pattern (that an Instance starts to expect things about Applications).

Steve

From: Alexander Tivelkov [mailto:ativelkov at mirantis.com]
Sent: Thursday, July 24, 2014 5:18 PM
To: OpenStack Development Mailing List (not for usage questions)
Subject: Re: [openstack-dev] [Murano]

Hi Steve,

Sorry I've missed this discussion for a while, but it looks like I have to add my 5 cents here now.

Initially our intension was to make each Murano component "self deployable", i.e. to incapsulate within its "deploy" method all the necessary actions to create the component, including generation of Heat snippet, merging it to the environment's template, pushing this template to Heat and doing any post-heat configuration if needed via Murano Agent.

That's why the deploy method of NeutronNetwork class is doing $.environment.stack.push() - to make sure that the network is created when this method is called, regardless of the usages of this network in other components of the Environment. If you remove it from there, the call to network.deploy() will simply update the template in the environment.stack, but the actual update will not happen. So, the deploy method will not actually deploy anything - it will just prepare some snippet for future pushing.

I understand your concerns though. But probably the solution should be more complex - and I like the idea of having event-based workflow proposed by Stan above.
I even don't think that we do really need the deploy() methods in the Apps or Components.
Instead, I suggest to have more fine-grained workflow steps which are executed by higher-level entity , such as Environment.

For example, heat-based components may have "createHeatSnippet()" methods which just return a part of the heat template corresponding to the component. The deploy method of the environment may iteratively process all its components (and their nested components as well, of course), call this createHeatSnippet methods, merge the results into a single template - and then push this template as a single call to Heat. Then a post-heat config phase may be executed, if needed to run something with Murano Agent (as Heat Software Config is now the recommended way to deploy the software, there should be not too many of such needs - only for Windows-based deployments and other legacy stuff).


--
Regards,
Alexander Tivelkov

On Tue, Jul 22, 2014 at 2:59 PM, Lee Calcote (lecalcot) <lecalcot at cisco.com<mailto:lecalcot at cisco.com>> wrote:
Gents,

For what it’s worth - We’ve long accounting for “extension points” within our VM and physical server provisioning flows, where developers may drop in code to augment OOTB behavior with customer/solution-specific needs.  While there are many extension points laced throughout different points in the provisioning flow, we pervasively injected “pre” and “post” provisioning extension points to allow for easy customization (like the one being attempted by Steve).

The notions of prepareDeploy and finishDeploy resonant well.

Regards,
Lee
  [cid:image001.png at 01CFA7E1.91270F10]
    Lee Calcote
    Sr. Software Engineering Manager
    Cloud and Virtualization Group

    Phone: 512-378-8835<tel:512-378-8835>
    Mail/Jabber/Video: lecalcot at cisco.com<mailto:lecalcot at cisco.com>

    United States
    www.cisco.com<http://www.cisco.com>

From: Stan Lagun <slagun at mirantis.com<mailto:slagun at mirantis.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 22, 2014 at 4:37 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] [Murano]

Hi Steve,

1. There are no objections whatsoever if you know how to do it without breaking the entire concept
2. I thing that deployment workflow need to be broken to more fine-grained steps. Maybe instead of single "deploy" methdos have "prepareDeploy" (which doesn't push the changes to Heat), "deploy" and "finishDeploy". Maybe more/other methods need to be defined. This will make the whole process more customizible
3. If you want to have single-instance applications based on a fixed prebuild image then maybe what you need is to have your apps inhertir both Application and Instance classes and then override Instance's deploy method and add HOT snippet before VM instantiation. This may also require ability for child class to bind fixed values to parent class properties (narrowing class public contract, hiding those properties from user). This is not yet supported in MuranoPL but can be done in UI form as a temporary workaround
4. Didn't get why you mentioned object model. Object model is mostly user input. Do you suggest passing HOT snippets as part of user input? If so that would be something I oppose to
5. I guess image tagging would be better solution to image-based deployment
6. Personally I believe that problem can be eficently solved by Murano today or in the nearest future without resorting to pure HOT packages. This is not against Murano design and perfectly alligned with it


Sincerely yours,
Stan Lagun
Principal Software Engineer @ Mirantis

On Tue, Jul 22, 2014 at 8:05 PM, McLellan, Steven <steve.mclellan at hp.com<mailto:steve.mclellan at hp.com>> wrote:
Hi,

This is a little rambling, so I’ll put this summary here and some discussion below. I would like to be able to add heat template fragments (primarily softwareconfig) to a template before an instance is created by Heat. This could be possible by updating but not pushing the heat template before instance.deploy, except that instance.deploy does a stack.push to configure networking before it adds information about the nova instance. This seems like the wrong place for the networking parts of the stack to be configured (maybe in the Environment before it tries to deploy applications). Thoughts?

----------

The long version:

I’ve been looking at using disk-image-builder (a project that came out of triple-o) to build images for consumption through Murano. Disk images are built from a base OS plus a set of ‘elements’ which can include packages to install when building the image, templatized config file etc, and allows for substitutions based on heat metadata at deploy time. This uses a lot of the existing heat software config agents taking configuration from StructuredConfig and StructuredDeployment heat elements.

I’m typically finding for our use cases that instances will tend to be single purpose (that is, the image will be created specifically to run a piece of software that requires some configuration). Currently Murano provisions the instance, and then adds software configuration as a separate stack-update step. This is quite inefficient since os-refresh-config ends up having to re-run, and so I’m wondering if there’s strong opposition to allowing the object model to support injection of software configuration heat elements before the instance is deployed.

Alternatively maybe this is something that is best supported by pure HOT packages, but I think there’s value having murano’s composition ability even if just to be able to combine heat fragments (perhaps in the drag & drop manner that was briefly discussed in Atlanta).

Thanks,

Steve


_______________________________________________
OpenStack-dev mailing list
OpenStack-dev at lists.openstack.org<mailto:OpenStack-dev at lists.openstack.org>
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev


_______________________________________________
OpenStack-dev mailing list
OpenStack-dev at lists.openstack.org<mailto:OpenStack-dev at lists.openstack.org>
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/20140725/08cef4cb/attachment.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: image001.png
Type: image/png
Size: 1773 bytes
Desc: image001.png
URL: <http://lists.openstack.org/pipermail/openstack-dev/attachments/20140725/08cef4cb/attachment.png>