Compatible Group metric tables and summary

[carlabramsson]

Hello!

Consider this:

Group Members Reporting: 20
Name: Currently Active Sessions
Alerts: 0
Min: 359
Max: 1,189
Average: 735.9

The values are from a compatible group of 20 Tomcat 6 servers. Problem is that I think one value are missing, where is my summary? :)
This haven't struck me until today (when we are live doh!) but alot of our metrics for everything from cache hits to sessions is something that you want to know across the cluster. Not just the average / server, min value and max value.

I looked at the source (of RHQ) and there is a metric for summary but it's not populated (tried to add it to the rhq/common/monitor/tables.xhtml).

So my questions are:

1) Any easy way to patch in the summary? Shouldn't be that hard...

2) Are there any plans to incorporate the summary?


Also, is there any easy way to build a custom page that displays metrics from multiple resources? We have a very visual page (built on internal bad stuff and not JMX-based) today that displays some red and green dots. A client side api (Seam Remoting?) would be very nice for this. Create a xhtml-page, upload it onto the server and go.

[mazz]

I don't understand what you mean when you say "summary". Can you be more specific what you mean?

As for customizing the UI, there have been many discussions and requests for this - we haven't implemented anything yet because we aren't sure what we want to do nor have we had time to do something yet.

Along similar lines, we have been investigating and prototyping a remote API solution (web services, remote EJB, CLI) that would allow you to write your own webapp/client UI. We have most of the CLI checked into SVN and we have some of the webservice impl checked in - but none of it is fully wired end to end.

For more info:

http://support.rhq-project.org/display/RHQ/Design-CLI
http://support.rhq-project.org/display/RHQ/Design-API

[carlabramsson]

Regarding summary, I made a quick mock-up:

http://files.me.com/carl.abramsson/khnx9k

The feature I'm looking for could be used like above. It shows the "summary" of the total number of active sessions for 20 webapplications deployed in tomcat setup as a compatible group.
In the example above I just took Max*20, Min*20 and Average*20, but in i a real world scenario this would not do ofc. Jopr would need to take activeSessions on ServerA + activeSessions on ServerB + ... for each measurement to get the actual "cluster" (or compatible group) number.

Do I make any more sense now? :0)

[mazz]

I see. You want the aggregated "sum" (http://www.thefreedictionary.com/sum) of the metrics across the group, along with the aggregated average as well as the min and max.

That is interesting. The important thing to remember here is that for most metrics that I can think of, showing the sum would actually be meaningless (e.g. how does the sum of all "Free Memory" metrics across a group of UNIX boxes help? What does the sum of all "CPU Load" metrics mean?)

In this one case, the sum actually would be helpful ("Total Number of Active Sessions").

I'm thinking perhaps we can enhance the plugin descriptor metadata to be able to flag a particular metric as having meaningful sums across a group of resources. Our UI can then show you the columns you want, but only for the metrics where it makes sense ("Total Database Connections" would also be a good one to aggregate sums for - on datasource connection pool resources).

you should write this up in an RHQ JIRA - http://jira.rhq-project.org/browse/RHQ

[ips]

Mazz-

Or we could automatically compute group sums for metrics with units="none". Or we could have a computeAggregateSums flag on metric defs, and default that flag to true for metrics with units="none" and to false otherwise. Or we could just do it for all metrics, and let the user decide if the sum is meaningful to them or not.