Strange jump in the graphs

[evalalos]

Gatling version: 3.10.5
Gatling flavor: java kotlin scala javascript typescript
Gatling build tool: maven gradle sbt bundle npm

Hello everyone, I have some strange jumps in the graphs and cannot understand why.
I have ‘main’ scenario with 4 ‘sub-scenarios’ added in the setup method with the injection open model.
In all of them I also have rampUsersPerSec and constantUsersPerSec methods.
So I want to have constant number of users arriving in the system and some of them to arrive during the certain ramp-up time.
So here is the example:
RAMPUPTIMESECONDS = 800 (should be around 13 mins (used for rampUsersPerSec))
EXECUTE_FOR_MINUTES = 5 mins (used for constantUsersPerSec)

NUMBER_OF_CLIENTS_PER_MIN is set to 10
Percentages are set with different values for different scenarios (eg. 30, 50, 20)

//calculates percentage of clients per second from total clients per minute
public static double getPercentageOfClientsPerSec(Double clientsCountPerMin, Double percentage) {
double percentageOfClientsPerSec = (clientsCountPerMin * percentage) / 100 / 60;
System.out.println("Percentage of clients per second: " + percentageOfClientsPerSec);
return percentageOfClientsPerSec;
}

I will attach a few graphs from a different runs. Also attached the simulation.

Any idea why I have that big jumps?

This started to happen few weeks ago.

Update:
I want to have it like this (or similar):

[evalalos]

@slandelle any thoughts on this? PS: Updated to the latest version

[slandelle]

I suspect you don’t really have an issue.
What you have is active users, not concurrent users, cf Open-Source
What you’re seeing is probably merely users lifetime spanning over a 1s bucket or not.

If you want more help, you have to provide a reproducer as requested.

[evalalos]

If that is the case, why I don’t have it more often in the graph, but only once?

The strange thing is that jump happens always at the end of the ramp-up time.
Providing two graphs from two different runs.

[slandelle]

Again:

If you want more help, you have to provide a reproducer as requested.

[evalalos]

It is a bit hard to provide everything since there are a lot of code, logic and dependencies.
I wanted to hear some ideas and thoughts.

[trinp]

did you check your response time during that time ?
If the response is long, some user may alive longer than expected

[evalalos]

Good idea, will check it now.
Thanks!

[trinp]

Once you figure it out, please share so that people know what to check if they stumble on the same behavior please
I’m also curious the root cause of this

[evalalos]

Of course i will share.
I think I’m close to understand what’s happening.

Btw, thank you for your idea, but the problem is not there.