juju-solutions / bigdata-kpi-scripts

KPI scripts (prometheus-pushback subordinate)
0 stars 4 forks source link

block a particular IP? #2

Open kwmonroe opened 7 years ago

kwmonroe commented 7 years ago

@ktsakalozos, is it possible to block stats from a particular IP? The latest big data KPIs show a huge spike for the hadoop-spark bundle. I've determined it's coming from the QA team and know the IP. Is it possible to block stats from this IP?

I think the answer is "no" because it seems we parse store data from https://api.jujucharms.com/charmstore/v5/{}/meta/stats. That feels too late to exclude a particular source, but I wanted to raise this issue in case there's another way to block an ip.

ktsakalozos commented 7 years ago

I think the way not to show up on the dashboards is to use a juju model-config -m ${MODEL} test-mode=1 before deploying. Would that be possible?

kwmonroe commented 7 years ago

Negative @ktsakalozos. QA is fetching the bundle archive (which is what we need to exclude), then they modify the local bundle to massage bits relevant for their restricted env (like adding proxies, etc). The models are test-mode=1, so we the actual deployment stats are already excluded -- it's the fetch of the bundle itself that is causing the spike.

I've talked with QA and they plan to drop their fetch from 4/day down to 1/day, so that'll help lessen the noise, but as long as we generate KPIs from charm store download stats, I think we're stuck with dirty data.