Hi, Just for the list to have it archived. We did briefly discuss this on irc. The changes have already been pushed.
On Fri, 2021-02-05 at 18:24 +0100, Mark Wielaard wrote: > On Thu, 2021-02-04 at 20:45 -0500, Frank Ch. Eigler via Elfutils- > devel wrote: > > - New prometheus metrics to track filesystem freespace, and more > > details about some errors. > > This part looks fine. > > Maybe this is a general question about exposing metrics. But does this > help people who want to do denial of service attacks? I mean by seeing > what effect some queries might have on freespace/errors? > > Not a showstopper, but maybe something to think about before > encouraging people to expose their metrics? yeah hypothetically, and yet at the same time one needs that info more to detect problems early > > - Frequent checking of $TMPDIR freespace, to trigger fdcache > > emergency flushes. > > This too looks fine. > > > - Switch to floating point prometheus metrics, to communicate > > fractions - and short time intervals - accurately. > > You move all metrics to use doubles. Including the metrics that use > integral values using inc_metric (the threads, _total and _count ones). > I cannot tell how that looks on the Prometheus side. Isn't it better to > keep those as integrals? Aka maybe inc_metric shouldn't call through to > add_metric? Or maybe have two metric maps, one for counts and one for > values/durations? doubles are the native type for prometheus consumers, the textual form doesn't really matter (and ieee doubles propagate 40+bit integers exactly so that's okay) might tweak the formatting to add all the digits of precision rather than the c++ default, easily > All code changes look fine. Please feel free to push if you think my > questions above are a little silly :) The actual changes seem to do > what you intended. Cheers, Mark
