How to identify when application breaks in load testing using vsts?
Demir last edited by user
i'm having hard time understanding the load test result in vsts to find when the application breaks under load.
Is there any indications to find it. I searched about this in google, asked in stack overflow but i dint get any clear answer about it.
Let me start by saying that analyzing the results of performance tests is a complex exercise. You need to have some understanding of how the system works, what parts of the system use up what resources (CPU, Memory, Disk I/O, Network Bandwidth, etc, etc). You also need to understand the architecture of your app and monitor all of the applicable systems and processes. It is often a good idea if you are not an expert, to enlist the help of an expert while analyzing the results.
There are many ways that a stress/load test can "break" your application. The obvious way would be that whatever process is hosting the application will crash. This is a little tricky to discover if it is an IIS application pool because IIS will automatically restart it for you. Often if this happens you'll see a rise in memory and cpu usage on the system, then it will bottom out for a bit, then begin climbing back up. This is usually an indication that something went horribly wrong. The fact that IIS recovered is great, but it usually means at the very least that all of the current sessions were immediately expired and can have other unintended consequences.
Another way could be that dependent services, processes or systems crash in a similar way and may or may not restart automatically. You have to ensure you are monitoring each of these.
Another way could be that even though the process has not crashed, something internally has met some limit or threshold, or there is a bug in concurrency or some other defect preventing the application from behaving in the desired way. This will usually show up in results as errors being returned in the responses from your web server, 500 errors are the most likely so you will need to validate the results coming in.
Another way would be that even though nothing crashes, the system slows to a crawl. You will need to determine an appropriate threshold that various activities need to fall within. If a response that takes