Update go_gc_duration_seconds HELP text

[thinkeridea]

# HELP go_gc_duration_seconds A summary of the GC invocation durations.
# TYPE go_gc_duration_seconds summary
go_gc_duration_seconds{quantile="0"} 1.1717e-05
go_gc_duration_seconds{quantile="0.25"} 1.5126e-05
go_gc_duration_seconds{quantile="0.5"} 1.9535e-05
go_gc_duration_seconds{quantile="0.75"} 4.3568e-05
go_gc_duration_seconds{quantile="1"} 0.000384508
go_gc_duration_seconds_sum 669.380082897
go_gc_duration_seconds_count 8.186899e+06

How these data should be used to facilitate analysis of GC performance.

I use the following queries in grafana to get graphs. I can't get effective information or hints from them. I can't understand the average and maximum time-consuming of gc, and the proportion of the maximum time-consuming.

avg(irate(go_gc_duration_seconds_sum{instance=~"$instance",project=~"$project"}[$interval]))
avg(go_gc_duration_seconds{project=~"$project", instance=~"$instance"}) by (quantile)

If possible, please explain how bitmap data can be queried to obtain graphs and how to analyze to obtain valid data, and use the above data as an example.

[beorn7]

Those metrics are coming directly from the Go runtime. For a better understanding of them, I recommend asking in relevant Go forums.

For Prometheus related questions, it makes more sense to ask them on the prometheus-users mailing list rather than in a GitHub issue. On the mailing list, more people are available to potentially respond to your question, and the whole community can benefit from the answers provided.

[bboreham]

I had the same question; this page is the top hit on Google for go_gc_duration_seconds_sum and I could not find any other page which answered the question. I had to read the code.

So, for the benefit of future visitors: the key insight is that this metric is the pause time, and since Go version 1.6 pause times have been miniscule - a few hundred microseconds max, which won't impact you unless you have seriously low latency requirements. For ordinary folks there is nothing much to be learned from this metric.

The Prometheus metric has a misleading name now; the duration of GC can be several seconds, but almost all of it runs in the background so doesn't directly impact execution.

[beorn7]

We could update the HELP text for the metric.

Reopening and renaming this.