Thanks for the input. Yes, in the statsd world, these are what I would have called gauges. HIstograms might be nice, but to get started, just the raw gauges will be a useful improvement over what we have now, so I figure I'd start with that. And, yes, I expect I'll implement this in some python scripts launched by cron under the toolforge jobs framework.
So, I guess if I wanted to do this on the command line, I would do: echo "some_metric 3.14" | curl --data-binary @- http://prometheus-pushgateway.discovery.wmnet/??? where the ??? is the name of my job. Do I just make up something that looks reasonable, or is there some namespace that I get allocated for my metrics? > On May 4, 2025, at 3:07 PM, Federico Leva (Nemo) <nemow...@gmail.com> wrote: > > Il 01/05/25 20:17, Roy Smith ha scritto: >> I want a graph vs time.. Which is what statsd/graphite was good at, so I >> assumed Prometheus would also be good at it. Why is this silly? > > It's not silly at all! If you use standard Prometheus metrics and some > labels, you can later also get some basic statistical analysis for free on > Grafana. > > What you described is called a Prometheus exporter. It would take the raw > data (from the MediaWiki API?) and output the metrics in Prometheus format. > You can hand-craft the metrics even in bash, but probably something like > Python or Rust where you have both MediaWiki and Prometheus libraries will be > easiest. > > The pushgateway is the traditional solution for a batch job like this. I > don't know how authentication etc. is handled in WMF though. > > The metrics you described are mostly gauges. For things like the time spent > sitting in queues, you may want a histogram (so you can calculate e.g. the > 75th percentile or the longest-waiting proposal). This is definitely best > done with a Prometheus library (but make sure to manually set the buckets to > some reasonable intervals, probably in terms of hours and days, otherwise you > might get some unhelpful defaults starting from ms). > > https://www.robustperception.io/how-does-a-prometheus-histogram-work/ > https://prometheus.io/docs/practices/histograms/ > > Best, > Federico
_______________________________________________ Wikitech-l mailing list -- wikitech-l@lists.wikimedia.org To unsubscribe send an email to wikitech-l-le...@lists.wikimedia.org https://lists.wikimedia.org/postorius/lists/wikitech-l.lists.wikimedia.org/
