CPU Mismatch between Tagged Web Request and its purepaths
Hi,
Our team noticed recently that when we open Tagged Web Request (lets call it TWR) dashlet, and pick a specific TWR and drilldown to its purepaths, the CPU Sum of TWR may not exactly match the combined CPU Sum of all the purepaths it drilled down to. This is happening for some of the TWR and not all of them.
On the other hand, for each TWR we have, the CPU Sum always exactly matches combined CPU Sum of all its Web Requests, which is good.
Similarly, if we drill down from purepaths (selecting all purepaths) to "methods" dashlet, the combined CPU Sum of all the methods exactly match that of the purepaths. This is also good.
So, it is the mismatch between TWR and Purepaths CPU Sum">CPU Sum which is bothering us. Some of the differences are huge starting from 400 ms going up to 4 sec.
We could not find any specific reason for this.
Can someone please clarify?
Regards,
VM
Vaibhav,
I'm don't understand the distinction you're making between "a specific tagged web request" (first paragraph) and "each tagged web request" (second paragraph).
Could you include a screenshot or two to illustrate the problem?
-- Graeme
Hi Graeme, I work with Vaibhav (different time zone). Here are some more details. We're using jmeter to push some traffic to our app, and we tag the requests in jmeter. So here's what the TaggedWebRequest dashlet looks like after we've stored our results. We ran 20 loops of this particular test.. We'll focus on the ADC_CardActivity transaction. You can see from this that dynaTrace is telling us that the avg / total CPU for this web request is 103.73 / 2074.59.
If I right-click this line and choose to drill down into purepaths, I see the following. Here are our 20 purepaths. Just eyeballing the CPU numbers, they are about 300ms on average, which is of course much higher than we saw in the TaggedWebRequest dashlet (the actual CPU total for all 20 purepaths is about 6109ms). If I drill down from here into the Methods dashlet, the CPU numbers agree with the purepaths.
So it would seem that the purepaths include a lot of activity over and above what is in the web requests. But we can't figure out what that is. If I select all the purepaths and then drill down into the WebRequest dashlet, I see numbers that agree with the TaggedWebRequests dashlet.
I can't drill down from WebRequests into Methods, so I can't compare that way. Just in case, I added a filter to the Purepath dashlet to make sure we're limiting the display to a single Agent, but that made no difference. We're stumped.
Vaibhav and Glenn,
Thanks for the additional data. The CPU sum shown in the Tagged Web Requests dashboard is the CPU sum of the first tier, not the whole PurePath. You can verify this by drilling down to PurePaths, selecting them all and then drilling down to Agent Breakdown.
If you can verify this on your system, I'll see if I can chase down whether this is a bug or works-as-designed.
-- Graeme
Thanks, Graeme. If I drill down to AgentBreakdown, it just shows that single agent, and the CPU number agrees with the Purepaths dashlet. If I drill down to TransactionFlow from either Purepaths or Web Requests, the result is the same for both. So I'm not sure if there's a tier that's included in the purepath but not the web request, and if so, how I would identify what it is and what's happening in it.
Glenn,
Is there any chance you can send us a PurePath so we can take a look at it?
If not, you might consider opening a support ticket.
-- Graeme
Glenn,
The Tagged Web Request dashboard is taking the CPU sum value from the CPU total of the first thread. You can see what's going on in the PP dashboard:
This is correct – Dynatrace is pretty consistent is using "total" to mean "including children" and "sum" to mean "all threads". You can find the remainder of the PP CPU sum in the other threads lower down in the PP.
The question is, why is the Tagged Web Requests dashboard displaying the value for the first thread?
Could someone from R&D say whether this is a bug or works-as-designed, please?
-- Graeme
Ok, this is interesting. The PurePath contains 3 asynchronous threads, which are included in the Purepath's CPU Sum, but not in the Web Request's (or Tagged Web Request's) CPU Sum. This is great information...it explains the sporadic differences we have been seeing. Maybe the CPU Sum column for the Web Request and Tagged Web Request dashlets should be renamed to CPU Total?
Glenn,
It's a bug. If you'd like a patch, you'll need to submit a support ticket.
Otherwise, you can wait for the fix to be rolled out in a regular release. If you post here in a week or so, I can probably let you know when.
-- Graeme
Thanks, Graeme. There's no urgency, now that we understand the issue. What exactly is the fix? Is it just labeling, or the calculation itself?
Glenn,
I expect they'll fix the calculation – there's no precedent for displaying the current value anywhere other than the PP tree.
The issue also applies to the ordinary Web Request dashboard, so you got a twofer!
-- Graeme
Thanks Graeme.
Yes, we noticed about WR dashboard also. In fact, it might be how DT consolidates and shows data at TWR level !
So, until this fix is available, we need to add CPU Sum of all Purepaths belonging to a WR/TWR to get a more correct measurement of CPU for that WR/TWR.
Is that correct?
-- Vaibhav
