Hi! i’ve been been working a while with k6 using output-prometheus-exporter and found something that i think it might be an error… but honestly i dont know if it is expected to work that way. so… before filling up an issue in gh i prefer to ask here.
I ran some test on a website and, while trying to understand http_req_failed metric, started to take a deeper view on http_reqs metric.
Among its labels i focused on “name”(the requests url) and “status”. I found that when responses are 200’s it properly starts from zero and count OK requests. Great! Then when there’s a 500, instead of creating a counter with status=500 from zero (as i expected) it starts with the value from 200’s metrics at that time.
For clarity, there’s a sample of a prometheus console query.
| metric | value |
|---|---|
| k6_http_reqs{method=“GET”, name=“<endpoint_1>”, proto=“HTTP/2.0”, scenario=“default”, status=“200”, tls_version=“tls1.2”} | 21 @ 1642424336.711 |
| 28 @ 1642424341.125 | |
| 43 @ 1642424347.734 | |
| 53 @ 1642424353.064 | |
| 69 @ 1642424359.359 | |
| 84 @ 1642424365.442 | |
| 101 @ 1642424371.087 | |
| 122 @ 1642424377.045 | |
| 131 @ 1642424378.63 | |
| 150 @ 1642424383.799 | |
| 165 @ 1642424387.92 | |
| … | |
| k6_http_reqs{method=“GET”, name=“<endpoint_1>”, proto=“HTTP/2.0”, scenario=“default”, status=“500”, tls_version=“tls1.2”} | 147 @ 1642424383.199 |
| 462 @ 1642424443.447 | |
| 733 @ 1642424489.655 | |
| 855 @ 1642424513.894 | |
| 1444 @ 1642424608.739 |
You can see that status=500 counter starts from 147, and if we look at the timestamps it fits right between the 131 and 150 values of status=200, and it seems to take that value as the initial. Then… when another 500 happens same occurs, it copies the 200’s and continues from there.
Also, if i count total requests made by endpoint… those with statuses 200 and 500 doubles the ones with only one status label while it should be in avg equal with all other endpoints.
Shouldn’t those counters start from zero and count only the responses by status separately? Am i missing something?
Thanks in advance!
and thank you for this question! This looks like a bug 
I created an issue here: