Strange behaviour of the virtual users

Hi.

I am having an issue in my tests. Virtual users are active, but they stop sending requests whithout a pause action, as can be seen in the figure bellow. To check if the requests really stoped, I used wireshark to capture the trafic and I confirm that the trafic goes down like the chart of number of requests per second. First I thought it was a problem in my scenario, but I executed the simple scenario bellow and I got the same problem. Then, I tried to decrease the number of users to check if there was not a performance problem of my computer, and the behavior persisted.

var snc = scenario(“simple”).exec(http(“test”).get(url))

setUp(snc).protocols(httpProtocol).maxDuration(10 minutes)

I changed the base url and I tested in another computer, but the virtual users stop sometimes and I don’t know why.

After all these tests, I thought this can be a bug in Gatling.

I used the version 2.1.7 and the last snapshop 2.2.0 too.

Did anyone have a similar problem?

Thanks.

My2cents:

  • full GC
  • C2 kicking in
    You’ll have to monitor your JVM (VisualVM or enable GC logs) and enable the compiler logs.

See https://wikis.oracle.com/display/HotSpotInternals/LogCompilation+overview

Cheers,

I followed your advise, but I did not see nothing strange in the log files. I put them attached. The number of requests per second is shown na figura abaixo.

I discarted lack of memory issues, because some of my tests ran with this configuration -Xms8G -Xmx8G -Xmn4G in a machine with 16 GB of RAM.

If you find something in the files or have another ideia, please, let me know.

Thanks.

gc.xml (1.26 KB)

compiler.xml.zip (3.04 MB)

Could you run with -XX:+PrintGCDetails so we get the GC times, please?

GC log file is attached.

Do you need the compiler log?

gc.xml (3.71 KB)

At what time did you see a pause occur when you got this log?
I don’t see anything that bad in the gc logs.
Are you sure you don’t have another process that might be suddenly eating all the CPU?