Open douglas-reid opened 3 years ago
for people installing prometheus with helm,
--set server.global.scrape_interval=15s
makes both the kiali 1min and most grafana istio dashboards work (compared to the 1min default of prom)
this being said 15s scrape is kinda frequent so indeed it'd be ideal that grafana dashboards adapts to scrape rate rather than the other way around (given prom doesn't consider giving "no data" a bug https://github.com/prometheus/prometheus/issues/8001 )
We should add documentation around the interplay of scrape intervals and rate ranges to our dashboard docs.
In helping troubleshoot dashboard issues today, we stumbled upon a "gotcha" of sorts with the dashboards / Prometheus. It looks like the default helm installs of Prometheus ship with a
1m
scrape interval. This is problematic, as our dashboards use a defaultrate
range of1m
. This means that the grafana queries into Prometheus will return no data (asrate
requires two data points). It looks like the general recommendation forrate
range configuration is 4x scrape intervals.At the least, we should have a warning in the docs about this very issue (and probably also in the appropriate FAQ).
It would be ideal to allow dashboard wide configuration for rate intervals. Grafana exposes the interval as a param, and that could be useful to adopt.
/cc @ldemailly @howardjohn