RHQ and performance/load test

[cristiano.nicolai]

Hi guys,

I'd like to share some opinion with you. I know that the primary goal for RHQ is monitoring but I think that this tool can be be very useful for perfomance and load tests as well.  The scenario that I'm thinking is to test some Java application inside JbossAS, all metrics needed are already collected by RHQ like VM stats, pools, threads and much more. So, if we could integrate with some tool like JMeter or Grinder (reading test metrics/generating graphs) and create a kind of snapsnhot of the selected resources during testing execution, will be a powerfull solution for performce improvement. I know that RHQ is extensible enought to do almost everything so I would like to know if there is some tought in this way.

cheers

[pilhuhn]

Christiano,

this sounds like writing plugins that can read data from Grinder or JMeter and show them graphically?

What exactly do you mean with "snapshots"?

Would you be interested to work on this?

  Heiko

[cristiano.nicolai]

Hi Heiko,

Yes I was thinking as a plugin to read test data and have all the information that I need (test data and environment metrics) in just one place/tool. So what a mean by snapshot is somehow aggregate all metrics that occurred during the test execution. Making possible to see something like test Response Time and server CPU%, maybe not in the same graph but at least creating a view like a dashboard where I can select the ones that I want.

Another thought  is that maybe we can reuse the agent infrastructure to run the tests. I'm not sure about this idea, if it is valid or not but basically we could use the agents to execute the tests simulating many clients.

I'm really interested in helping you guys

[pilhuhn]

Cristiano,

Yes I was thinking as a plugin to read test data and have all the information that I need (test data and environment metrics) in just one place/tool. So what a mean by snapshot is somehow aggregate all metrics that occurred during the test execution. Making possible to see something like test Response Time and server CPU%, maybe not in the same graph but at least creating a view like a dashboard where I can select the ones that I want.

That should be possible if you have all the (numeric) data by just re-arranging and adding graphs to one monitor page. In RHQ4 it should/will (hopefully)  be possible to define a new dashboard with portlets of your choice, which could include those metrics.

Graphing of response time data is a different thing - we used to have some graphs, which were not really useful. I guess one would need a 3D plot to get useful graphs here. Perhaps you have an idea?

Another thought  is that maybe we can reuse the agent infrastructure to run the tests. I'm not sure about this idea, if it is valid or not but basically we could use the agents to execute the tests simulating many clients.

You can start running tests from the agent plugins via operations. So if you have an infrastructure with multiple servers (+agents) you can set up the plugin on all of those and then trigger a group operation so that they all fire their test requests onto the target server.

I'm really interested in helping you guys

Excellent :-)

Thanks

  Heiko

[cristiano.nicolai]

Nice, so I think that the first step could be related to the chart engine. I think that something like a new dedicated view just for graphs/reports would be nice where the user could select metrics that he wants to cross, some kind of configuration and them save that with many portlets. I know that you guys are in process to change the UI to GWT and I also saw an idea about using JFreeChart or DOM+CSS. So, let me know exactly the requirements and them I can start some investigation.

Thanks again Heiko.

[pilhuhn]

Cristiano,

I think you should first subscribe to the rhq-devel mailing list (https://fedorahosted.org/mailman/listinfo/rhq-devel ) and bring your ideas to it - I can't comment on the implications of various graphing engines and the GWT usage.

We did have at one point (Developer Preview 1) have graphs that looked to me like being rendered directly on the DOM from within the Browser. While this 'stresses' the browser more, I think it also allows for greater flexibility like when you not only supply 60 data points (number in the old graphs), but e.g. 600 so that some drill in is possible.

Anyway, what I really wanted to say is: before selecting a technology, we should perhaps go one step back and define what we want/need for graphing - especially for Response Time data, where we have multiple dimensions: time (of day), url, min/avg/max response time per url. So I could imagine, that some 3d candle stick chart would be cool.

Or join us in the afternoon (Germany time) on irc.freenode.net in #rhq to discuss with the devs.

Thanks

  Heiko

[mazz]

There is already the start to a design wiki that talks about  different graphing libraries and some requirements.

See this: http://rhq-project.org/display/RHQ/Design-NewChartingLib

[pilhuhn]

Cristiano,

the wiki page Mazz has mentioned has been updated - along with some pointers to converstations in this email: https://fedorahosted.org/pipermail/rhq-devel/2010-October/000361.html

Is there anything where you need help helping us?

  Heiko

[cristiano.nicolai]

Thanks guys, I'll join the list to give my 2 cents