I'm working on an online reservation system (the details aren't really relevant to my issue). It's mostly developped in C# with framework 4.0. I have a DB which contains, among other things, the products and the availability of these products. The DB is accessed by a web service, the latter being queried by a web application, which in turn sends the information to be displayed to the browser. Nothing too fancy in term of architecture, but I don't have much experience with that kind of system. This projet is quite old; it's been in development for a few years and I just hopped in six months ago, having no web programming experience before.
I'm currently experiencing concurrency problems. If I open two browsers window (any browser), enter the same query in both windows and post on both pages as fast as humanly possible, I get unhandled exception in my web service. One of the client gets the data, the other receives an error. EDIT : Also, I tested using two different computers simultaneously, and I still could get the unhandled exception.
I was able to trace the problem back to a cache system that was implemented almost a year ago. There's is no locking mecanism, and the cache uses a Dictionary (which is not thread-safe). A few Stack Overflow questions helped me in pinpointing the problem and how I should correct it (this one and that one for example).
That being said, I'm trying to reproduce the problem without having to manually click in two browser windows and hoping to get the right timing to access the same cache simultaneously. The reason I'm trying to do that is that I don't like implementing a solution that should work when I can't reliably prove it actually worked. Also, there's a lot of places where the cache has been (badly) implemented, and I'm not sure I can test all those places by "guessing" the right timing manually.
So, I downloaded Apache JMeter, and using Fiddler2 hooked up to my two browsers, I could find all the HTTP requests I needed to create a Test Plan that sends exactly the same requests.
But to my dismay, JMeter couldn't reproduce the problem, even if I set my Thread Group to create 40 concurrent users! Each one of them receives the data correctly, and none triggers the unhandled exception I am getting by doing it manually. I know it's not a cookie issue, nor a difference in user-agent, as I tested with multiple browsers and with disabled cookies. As the HTTP requests were created directly from Fiddler2, it shouldn't be an issue with hidden fields or URL rewriting, and there's no dynamic field names or content involved. I know Fiddler2 has a "replay" function, but even that couldn't reproduce the problem. This baffles me to no end.
So, my questions are : why can't I reproduce the problem with JMeter? Is there a better tool than JMeter (which is quite complete for a free software) to do that? What are the browsers doing that Jmeter doesn't do (apart from rendering the page, which is really not the issue here)?
Any thought would be much appreciated!
