Apache CGI instrumentation feedback?

andreas.grabner@dynatrace.com: Any idea on how we could instrument this?

Hi

I have not done this before. I dont share the fear of the Native ADK having too much overhead. We are using the Native ADK in many installations with heavy load - overhead is never an issue as it is extremly lightweight.

Andi

Thank you Andi for your reply. I will forward your response to the customer and I'll let you know if they have any further question or comment about this.

Hi Andi -

Would you be able to expand on the comment "extremely lightweight":

• What is the actual overhead that you are seeing in most cases? 
• Do you have any data that you can share on this? 

Thank you!

The customer would like to have more details, especially about how to instrument a CGI application and how other customer did it in the past.

Do we have examples we can communicate of other customers instrumenting their CGI application with dynaTrace?

Eric: With extremely lightweight I mean that our API calls take the data our users capture and put them into a queue that gets send asynchronously to the dynatrace server. This is the same approach we use for our Java & .NET Agents. FOr Java & .NET we typically capture much more data than with the ADK as we have a lot of out-of-the-box sensors. As you know we never really have customers complain about the overhead for these technologies. As the ADK implementation typically captures less PurePath nodes the overhead of this is even lower than you would see in Java & .NET. I don't have any data on-hand right now but I will see if I can get anything from our engineering team on this

Christopher: I hope somebody else can chime in her. I am not aware of an actual implementation. Therefore I cant share any stories

hi Christopher,

so my assumption about this CGI environment is as follows: the CGI scripts get specific (or all?) HTTP requests from the apache and creates new HTTP requests which call then into .NET. I assume for every incoming HTTP request into apache/CGI one outgoing to .NET.

so if that's the case I'm wondering if you really want to use the ADK for this, as I see (at least) two options here:

1. you just take the "X-dynaTrace" header in the CGI script from the incoming HTTP request and set it on the outgoing one.
2. you really use the Native ADK in CGI, have to read the "X-dynaTrace" HTTP header, set the dynatrace tag, start a purepath, send events, create a link (all via ADK) and again set the "X-dynaTrace" HTTP header

option #1 is of course a lot easier, does not require the ADK and would continue the PurePath, but would not show you insights about the CGI scripts.

option #2 would - with the help of the ADK - in addition give you insight into the CGI scripts.

so if you're not after this additional insight into the CGI scripts, option #1 would be way easier IMHO.
however, if the CGI scripts are doing more complicated stuff, then the use  of the ADK might be necessary, anyway.

HTH,
Christian 

Thank you Christian for your reply: unfortunately, it looks like the intermediate layer written in C/C++ can potentially make multiple outgoing call for a single incoming web request so I don't think the 1st solution would work.

Option #2 would therefore be the only real option here but the customer is hesitant and would like to have to feedback on the overhead caused by the tool as, from my understanding, there is a new process spawned for each incoming web request and they're afraid that the dT agent initialization would take too long.