Figure out which datapoints (and how to measure them) we care about for load testing

mozilla-services / stubattribution-loadtests

0 stars 5 forks source link

Figure out which datapoints (and how to measure them) we care about for load testing #1

Open stephendonner opened 7 years ago

stephendonner commented 7 years ago

Just keeping a note here to help organize my notes/thoughts/questions about which metrics/datapoints we care about (and how/can we measure them) during load testing:

requests per second (RPS)
time to first byte (TTFB)
CPU(s) usage (running 'top' command vs...?)
memory usage on the server(s)
load average (ancient article: http://www.linuxjournal.com/article/9001)
disk I/O activity

stephendonner commented 7 years ago

@oremj - when you get a chance, can you please weigh in, here, on:

1) the above (and any additional metrics/things we should be testing/looking for) 2) and, as specifically as possible, which parts of the service, in particular (interactions, requests, return codes, etc.) we should be using and trying to push, hard, to break through caching, and stress-test the service?

Thanks!

oremj commented 7 years ago

I'm looking for req/s and response time.

Changing the attribution_code, product and lang parameters will suffice for cache busting.

chrismore commented 7 years ago

@oremj can we set baseline for req/s and response time that we can say "if these metrics are <= we are good to proceed"?

oremj commented 7 years ago

I mostly just need an idea of where we are, so I can scale appropriately.