Prometheus and Grafana

[ghost]

I was thinking of writing some documentation on Prometheus, exporters and Grafana and submit a PR. WDYT? The idea I suppose is to bypass the Locust UI. I think the node and stats exporter could be exampled. As you mention the swarm rate is go routines and we have the max rate built in via your flag.

[myzhan]

Any PR is welcome. In my daily work, I did similar things with Jenkins and Docker, and it's doable.

[ghost]

I was thinking of using the Locust --no-web switch and awk to parse console output and netcat to send the metric to the statsd_exporter. Pretty low-fi , I know - but I should be able to graph metrics in Grafana.

[myzhan]

Another option with the WebUI.

1. Start 100 users with a hatch rate of 10.

curl --data "locust_count=100&hatch_rate=10" "http://locust:8089/swarm"

2. Collect json stats every 2 seconds, like the ajax request.

for ((i=1; i<= 300; i++));
do
  curl -s "http://localhost:8089/stats/requests" >> locust-stats.json;
  echo "" >> locust-stats.json;
  sleep 2;
done

3. Stop the test.

curl -s "http://localhost:8089/stop"

You will get locust-stats.json, easier to parse.

[ghost]

Thanks for the suggestion but I am thinking the boomer rps flags and the locust user/ swarm rate could be confusing for the load tester, and I am leaning towards no interaction with the Locust UI. Although I could be heading in the wrong direction.

[ghost]

Yeah, Looks like a real hassle parsing stout or using tee. Getting broken connection and peer resets when using the Locust UI.

[ghost]

@myzhan Your technique works. Thanks.