On Wed, May 11, 2016 at 5:01 PM, Gary Brown <gbrown@redhat.com> wrote:

Hi

I would like to discuss how best to define the concept of an Application within the BTM project, how it relates to the existing concept of a "Business Transaction" and how it may link in with the concept of an application in Hawkular/ManageIQ in general.

Firstly, for those not familiar with Hawkular BTM, without configuration the project can capture and trace an invocation across multiple interconnected services. The information can be used to provide stats on the individual components used (endpoints, databases, ejbs, etc) as well as (in the next version) present a graph showing the relationships between the communicating services.

The project also enables "Business Transactions" to be configured - this enables a particular invocation of an endpoint to be labelled with a business transaction name (which is propagated to the fragments associated with the invoked services), and also perform additional processing on the messages, such as extract business properties, etc. Once a "Business Transaction" has been defined, it is possible to view high level stats about the complete "end to end" business transactions.

This is achieved by creating a configuration that identifies the initial endpoint for the business transaction, using a regular expression, and then a set of 'processors' that are used to perform various processing tasks within the scope of that business transaction.

We now have a requirement to identify the concept of an Application, and be able to present various stats about it. I believe that the Application can be viewed as orthogonal to the Business Transaction concept - so an Application represents the various operations that can be performed on a particular architectural component, whereas the Business Transaction represents a particular path through multiple Applications/Services, only using a subset of the operations/endpoints supported by the Application.

I quite agree here, except that I would map an application to *several* achitectural components (for instance N microservices).

Now if you have 2 logical applications (say e-commerce and brand website), a same architecture component may be shared (say user registration) in the 2 apps.

However the requirements are very similar - when we detect some activity on a particular endpoint we want to determine if it belongs to an Application, in the same way as we currently do with Business Transactions. And similarly, once classified as being associated with an Application, we may want to do some application specific processing on the data.

So here if an activity on a user registration endpoint is recorded, would it record for each segment with app it referred to ?

Therefore what I am considering is making the existing Business Transaction Configuration generic so that it is suitable for both tasks, with a simple classifier to indicate whether it relates to a Business Transaction or Application.

This will mean that the fragments of activity being recorded and sent to the BTM server could now potentially have two names, one for the business transaction (which is propagated across application/service boundaries) and one for an application name (which is not propagated).

Where appropriate, the UI could then be updated to allow the user to also filter information based on application - this may only make sense on the APM page, but could be considered if useful elsewhere. We may also want to make the current Business Transaction overview page more general to include both Business Transactions and Applications, and then have a different type of details page for applications.

The final point is how this may link into Hawkular/ManageIQ. Was thinking that where possible, it would be good if a deployment event could be used to trigger the configuration of the Application in Hawkular BTM - especially if it is possible to obtain the web context for the application (or JMS queue/topic), which can then be used to establish the regular expression. If this level of integration was possible, then no user defined configuration would be necessary to capture application information using the same name as in Hawkular - allowing an integrated view of the information to be made available.

Thoughts?

In general from the very high level view (ManageIQ), I would want to be able to define a logical notion of "application" for monitoring purposes, I would expect to be able to add "entities" to that logical view. It may just be a war file (and in the case of BTM, I would likely be interested by all entry points to that application) or it could be an "OpenShift Project" with all its dependencies (all docker containers + everything in it), or it could be a selection of MW servers...

In the end, the user should be able to dig into infrastructure details (and he already has that, he can see details about the infrastructure: a container, a host, a MW server (coming soon), a MW deployment (coming soon)...) + details on Business transactions but also have that synthetic view where he knows that all services that makes his online store are running fine, and have a few KPI such as the Apdex for all user-facing interactions, the N slowest URLs or DB requests...

(This definitely goes beyond BTM, and we likely need to simplify and work on iterations of those ideas (or better ideas))

That's what I dream of, as a user.

Thomas

Regards
Gary

_______________________________________________
hawkular-dev mailing list
hawkular-dev@lists.jboss.org
https://lists.jboss.org/mailman/listinfo/hawkular-dev