[openstack-dev] [Vitrage] questions about "relationships"

Rosensweig, Elisha (Nokia - IL) elisha.rosensweig at nokia.com
Wed Apr 20 08:44:55 UTC 2016


Hi Tony,

I agree. Looking forward for those challenges - should be an interesting development process.

Elisha

From: EXT Wang, Tony T (Nokia - CN) [mailto:tony_t.wang at nokia.com]
Sent: Wednesday, April 20, 2016 9:31 AM
To: OpenStack Development Mailing List (not for usage questions)
Subject: Re: [openstack-dev] [Vitrage] questions about "relationships"

Elisha,

Thanks for your detail explanation. I think I can understand how Vitrage works now. :)
Since Vitrage needs to read data from different data sources and build one Entity Graph, the elasticity of vitrage-grpah process will be a big challenge per my understanding. :)

Thanks again,
Tony

From: EXT Rosensweig, Elisha (Nokia - IL) [mailto:elisha.rosensweig at nokia.com]
Sent: Wednesday, April 20, 2016 12:26 AM
To: OpenStack Development Mailing List (not for usage questions)
Subject: Re: [openstack-dev] [Vitrage] questions about "relationships"

Hi Tony,

Glad to help :) . Regarding your questions from the last two emails:


1.       Yes, the Entity Graph (what you called the global graph), is generated regardless of what templates were defined. You can get this graph using the CLI command "vitrage topology show", and we are currently in the finishing stages of also having Horizon support for this that displays the graph.

2.       Yes, the template definitions have no direct impact on the Entity Graph. The impact on the Entity Graph is only via the template actions (see answer for next question)

3.       The "add causal relationship" will indeed add an edge to the Entity graph, between vertices already in the graph. Other actions, like raise_alarm, can cause vitrage to raise an alarm which will be added into the graph as well.

As a result, depending on the templates you add, the graph could indeed become very "dense" with edges. As Vitrage matures, it might also support many more types of relationships and entities, and the graph will grow accordingly.

We've given much thought to this issue, and have plans on addressing possible problems that this could cause. For example, the Entity Graph is currently in-memory, and we have plans for supporting also persistant-DB implementations, e.g., Neo4J, Titan.



As for your last question, yes - that is what the EvaluatorEventTransformer handles.



Keep those questions coming...



Elisha


From: EXT Wang, Tony T (Nokia - CN) [mailto:tony_t.wang at nokia.com]
Sent: Tuesday, April 19, 2016 5:37 PM
To: OpenStack Development Mailing List (not for usage questions)
Subject: Re: [openstack-dev] [Vitrage] questions about "relationships"

One more question, I suppose "EvaluatorEventTransformer" is used to handle events sent by ScenarioEvaluator, and add/delete vertex/edge to the "global graph" triggered by Actions. Is it right?

From: EXT Wang, Tony T (Nokia - CN) [mailto:tony_t.wang at nokia.com]
Sent: Tuesday, April 19, 2016 10:01 PM
To: OpenStack Development Mailing List (not for usage questions)
Subject: Re: [openstack-dev] [Vitrage] questions about "relationships"

Hi Elisha,

Thanks for your quick response. I think I got your points.
To double confirm, further questions are coming :):

1.       Regardless if there is template definition, Vitrage constructs a graph based on the data from plugins defined in datasources conf. Let's call it "global graph". Is it right?

2.       "entities" and "relationships" defined in template "definitions" won't add any new vertex and edge to the "global graph". They are only used to be referred  by "condition"/"action" in "scenarios" section. "condition" also uses them to construct another graph("ScenarioEvaluator._evaluate_and_condition"), let's call it "template graph". This template graph is used to  find occurrences of a template graph in the "global" graph("NXAlgorithm.sub_graph_matching").

3.       If so, which graph will be applied by the actions by ScenarioEvaluator?

For example, will "AddCausalRelationship" add "causes" edge to the "global graph" or "template graph"?

I suppose it should be "global graph". If so, it is possible the "global graph" will grow very huge someday?

Thanks,
Tony

From: Rosensweig, Elisha (Nokia - IL)
Sent: Tuesday, April 19, 2016 7:00 PM
To: Wang, Tony T (Nokia - CN); Weyl, Alexey (Nokia - IL); Afek, Ifat (Nokia - IL); Shamir, Ohad (Nokia - IL)
Subject: RE: [Vitrage] questions about "relationships"

Hi Tony,

Thanks for your interest in Vitrage! Any questions you might have, please send us. You may also send them to the openstack mailing list, and we will answer there as well.

Regarding your question:


*         In the templates, the definitions in the "definitions" section are used only by the scenario(s) within the template - the condition, as well as the action, might reference these.

*         These templates are used for business logic for the Scenario Evaluator, to detect patterns in the Entity Graph and perform subsequent actions.

*         On the other hand, the populating of this Entity Graph is done via Datasources. In the Transformer of each datasource, we define how we want to integrate the data received into the Entity Graph, what labels each edge (=relationship) will have etc.

For example, the InstanceTransformer will define that whenever we add an instance to the Entity Graph, the instance is related to it's host with a "contains" relationship. The template, on the other hand, states "IF there is a host in error AND this host contains an instance, THEN do XXX".

So indeed, we need the conditions in the template to use the same "language" as the transformers do, for example in using the labels. Currently, this mapping is implicitly assumed to be identical (i.e., the two use the same labels). In the future we will probably add a mapping configuration file, which maps the transformer values to those in the templates.

Is this clear?

Thanks,

Elisha

From: Wang, Tony T (Nokia - CN)
Sent: Tuesday, April 19, 2016 1:13 PM
To: Wang, Tony T (Nokia - CN); Weyl, Alexey (Nokia - IL); Afek, Ifat (Nokia - IL); Rosensweig, Elisha (Nokia - IL); Shamir, Ohad (Nokia - IL)
Subject: RE: [Vitrage] questions about "relationships"

Looks like "EvaluatorEventTransformer._create_neighbors" is used to create the edge for relationship.
Is so, is it overlapped to define the relationship between "nova.instance" and "nova.host" since  "InstanceTransformer._create_host_neighbor" should do the same work?

Thanks,
Tony

From: Wang, Tony T (Nokia - CN)
Sent: Tuesday, April 19, 2016 5:35 PM
To: 'alexey.weyl at nokia.com'; 'ifat.afek at nokia.com'; 'elisha.rosensweig at nokia.com'; 'ohad.shamir at nokia.com'
Subject: [Vitrage] questions about "relationships"

All,

I found Vitrage is an very impressive OpenStack project! :)

I have a few questions about the example:
https://github.com/openstack/vitrage/blob/master/doc/source/vitrage-template-format.rst#example-1-basic-rca-and-deduced-alarmstate
In this example, there is a "relationships" definition between nova host and nova instance:
- relationship:
                source: host
                target: instance
                relationship_type: contains
                template_id : host_contains_instance

My questions:

1.       Is "relationship" used to create an edge or not?

2.       I don't find the code for edge creation from "relationship". If not, the edge between "nova.host" and "nova.instance" should be created by  function "InstanceTransformer._create_host_neighbor". Right?

3.       If so, what's the purpose for "relationship"? Is it only used to parse the "condition" part in "scenarios"?

Thanks in advance,
Tony

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstack.org/pipermail/openstack-dev/attachments/20160420/fbff67bd/attachment-0001.html>


More information about the OpenStack-dev mailing list