Any way to count unique?
After some searching, I found some posts referring to inability to filter unique pure paths, is this still the case?
Trying to figure out how I can chart unique web requests
For me, that is a PP that is going to a specific URI and contains a request header value, I got as far as splitting by that unique value, but I want to chart the count of those unique splittings
Is there perhaps a lesser elegant way of doing it, knowing that we don't have ability to count unique?
Hi Roy
Can you tell me again what makes your requests unique? Is it the unique combination of a URL and an HTTP Header?
Hi Andreas,
A unique request defined by the 'cookie' http header value (or specifically the session id that i'll need to extract out of the header value). This represents a session that's been established on the server. The Web request is to a specific URL only, so I only need to filter for that URL.
Here's how it works:
- First visit, browser has no session cookie to send (i.e. no session cookie in the 'cookie' request header)
- Server creates new session and returns the new session cookie back to browser.
- Subsequent visits, browser will send the session cookie.
I am looking to track these session cookie creations to extract a count of unique visitors.
Two logical options to obtain this count would be to (A) count the number of occurrences of step 1, since not sending a session cookie would mean they are a first time visitor or (B) count unique number of session cookies sent in step 3.
First, I tried to do option B. Created BT with necessary filter, then split by session cookie value. It's as far as I could go because I wasn't sure how to count those splittings.
Second I gave up and tried to do option A, which is to count step 1. This means filtering for PP's that have that URI as well as no session cookie in the request. Creating a measure filter for "no session cookie in the request" proved challenging. I started by creating a 'Web request - Header Value' filter, with header value 'cookie' not contains value 'JSESSIONID'. So far so good. BUT when i tested it, it did not give me the expected PP. I inspected the PP, and realized that for some requests, if there are no other cookies to send, the 'cookie' header may not exist at all (perfectly normal). But looks like when filtering for 'cookie' not contains 'JSESSIONID', the BT does not include the case where 'cookie' does not exist at all 
So I went looking for a measure filter for not containing a header, but none exists.
Any other alternatives welcome 
I understand. So - typically this session handlign is done by your application server - correct? I assume you use something like a Tomcat, JBoss, ... If that is the case then you should look into the JMX Metrics that these app servers are exposing. Dynatrace already provides a list of pre-configured Measures for that. Open your System Profile -> Measures -> Create Measure and there expand Server Side Performance -> Agent-based -> Tomact. Look at "Manager Active Session" or "Manager Session Counter". These are values that Tomcat is exposing via JMX. If you are not using Tomcat but a different appserver then I am sure there is an equivalent to these counters
Andi
Thanks Andreas. I should have mentioned that JMX is what I am primarily using right now to chart active sessions. I was in disbelief that I had such a high number in a previous load test run (over 8k), and I was looking for a way to corroborate that number with another way of counting sessions
Anyway, thanks again.
Roy,
just saw you are from hybris . I've developed the hybris fastpack for monitoring hybris with dynatrace and doen most of the work with hybris.
Your example, counting number of active sessions is not uncommon and I can tell you from my experience with som hybris implementations that the JMX values you get from the application servers are correct and the best way to count unique sessions.
Likely in your loadtests you really created that many sessions because the loadtest threads didn't re-use the sessions. I've seen that before and it resulted in a not really meaningful test as too many sessions were created (e.g. a load test with "only" 100 users esulted in thousands of sessions as every request in a test created a new one). It was fixed by fixing the loadtest .
Another option that we investigated back then was to instrument the session creation in java itself to verify this but we found that the JMX values were correct anyway without the need of instrumentation.
The ultimate option would be to use User Experience Management , but that only really works well for real users not for load tests.
Reinhard
Hi Reinhard,
I've used your set of dashboards as the basis for my own Thanks a lot for that.
Went through the exact same thought process of tracking users. Yes, I am sticking to JMX for now. Pretty sure you are correct with your assessment on the high number of sessions. Something like JMeter might re-use the same session if loop count is forever, but I am using the synthetic testing (Gomez). I suspect it simply spins up a new test when the existing one ends, causing a new session to be created. I think with Gomez, 800 'virtual users' for 60 minutes ends up creating a zillion active sessions if your test is very short, since it will try to meet that count by starting a new test. Which there was a way to loop using the script recorder but no such luck. For my next opportunity, I will switch back to JMeter, but there's always the challenge of making it truly distributed!
Reviving this old thread Any of you guys know for sure how I can get Synthetic Load Testing to re-use the same sessions? i.e. the equivalent of looping in JMeter where if test ends, it should reuse that browser rather than spinning a new browser (and new session). If only there was a way to loop in test recorder....
Roy
I just stumbled accross this yesterday. There is a setting in the Dynatrace Recorder Script options that seems to do exactly that "Run script as repeat visitor" - only if used with the IE agent though.
Reinhard
OK, so I tried playing back my script with IE agent in the script recorder, to make sure all is good before scheduling a launch.
- With script recorder 5.x I get an error saying IE11 is not supported (even though the drop down says IE10???)
- Upgraded to recorder 6.x (latest) and my script errors out now saying execScript function not found. Arrgh
- I guess recorder tool is not backwards compatible. Re-recorded a new script in 6.x just as a test. Launch in IE agent. Hangs when trying to input search text in a search input box, whereas the other agents continue to run the test OK.
- Trying to stop the IE agent basically hangs the script recorder.
- Arrrrgh.
Have you tried playing back your script with IE agent? Don't know if it's just a recorder tool crappiness, or should i trust it should work when in synthetic load test?
The recorder can be tricky...
I'm not a scripting expert but I've played a bit with it.
I'd just give it a go running your script on the platform....and see.
For other script related questions I can recommend the synthetic forums.
