K6 VU user count off in Grafana

[eva2000]

@ned

Yes, but the VU iterations will also need to be more than a second long. k6 will hit whatever VU target you give it, but if the next target is lower, it will immediately try to decrease the actively running VUs. And if either your iteration duration or your gracefulRampDown option is super short, the 1-second measurement and emission of the vus metric might not coincide with the short peak of active VUs.

I see. I compared my previous ones with current and noticed that after I introduced a randoms sleep time between 1.5s to 4.5s, reported VUs are slight more accurate because of this?

before

iteration_duration.............: avg=1.85ms   min=555µs    med=1.15ms   max=36.28ms p(95)=5.13ms   p(99)=12.22ms p(99.99)=30.84ms  count=46674

after

iteration_duration.............: avg=2.01s    min=206.09µs med=2s       max=5.03s    p(95)=5s       p(99)=5s       p(99.99)=5.01s    count=2707351

@ned

If you don’t want to mess with these (and you might not be able to, depending on the test), you can also add a stage with the same peak target and a duration of a few seconds. Like this:

Sweet, thanks for the tip