java Socket - legacy system time isolated in the transaction flow
Hi,
We are using Dynatrace on a Java backend server that calls some external system through a java.net.Socket connection (which is a legacy system that has no dynatrace agent).
When I consult the purepaths that are produced, the Transaction Flow indicates that 100% of the time is spent on my java machine is not correct since most of the time is spent in the legacy system.
Is there a way to configure in the system profile, the fact that a call to the class java.net.Socket produces time that is spent on another tier ?
Thanks,
Here is one screenshot that shows the elements of the purepath :
Hi Nicolas
Which version of dynaTrace do you use? With dynaTrace 4.2 we introduced a feature that allows visualizing "External Calls" in the transaction flow.
Also - could you export one of these PurePaths so that we can have a closer look at it?
Thanks
I use dynatrace 4.2. Here is the attached : dynatrace_socket_legacy_time.dts
The feature you mention : an "External Calls" box in the transaction flow is exactly what I need. So perhaps there is an option to active or display it ?
hi Nicolas,
the feature 'external calls' only works for calls that we are able to detect (web services, HTTP calls, remoting, ...).
however, in your case it seems to be plain and proprietary java socket communication. in this case dynaTrace is not able to detect such calls.
so if you want to get to the time spent in this tier, i currently see 2 options:
- easy: you find methods on the java side, which are executed during the time of these calls, place sensors for them and work with APIs to get to that time. in this case the transaction flow stays the same, but the time is reflected in the API.
- not so easy: if possible, you inject an ADK agent into the legacy system (with source code instrumentation) and use the Java tagging ADK to tag the calls into the legacy system. in this case you would also get a complete transaction flow.
best,
Christian
Thanks for the answer, I thought that dynatrace could be able to identify that because a use of a Socket means by definition an external call and the time reading the stream from the socket is the time of the external system.
It could be perhaps an evolution for a next release of dynatrace.
Nicolas,
Nicolas,
actually reading from the socket doesn't necessarily mean time spent in the "other system". It could be a resource bottleneck in the calling system, in the called system or even in between. As Christian described the most accurate measurement you can get without having an agent in the called system and without ADK tagging is to define sensors on methods that are actually calling the external system and either use the API time or the execution time of that method as an indicator.
Maybe in your case placing a sensor on fr.msa.agora.s44accesgi.metier.gestionindividu.service.GestionIndividuImpl.* makes sense, as it looks like there is one or more services beeing called that you maybe want to distinguish.
Reinhard
basically that would be possible.
however, it would make no sense at all to show each single socket read/write as one external call. so we would have to collect multiple reads/writes and maybe aggregate based on destination IP or similar. and then we would have to make sure that we don't count times twice e.g. on socket and on web service. so there'd be a need for some kind of correlation and so forth ...
so it may be some future release feature, but no trivial for sure.
and maybe we also find another (e.g. network based) approach ...
best,
Christian
Community Member of the Month
December 2018
Announcing Dynatrace's Community Member of the Month for December 2018, Sebastian K.! Click here to read more! Employee Member of the Month
December 2018
Announcing Dynatrace's Employee Member of the Month for December 2018, Joshua P.! Click here to read more! 
