I have built a quick demo that shows that the labels are `inherited` by the spawning goroutine for profiling. (I clearly hit the golang-nuts button too quickly for this question)
https://play.golang.org/p/SbgtWB1fIKB Here we have a fairly pointless function which calculates Fibonacci a number of times. Once as a straight function call and once in a goroutine. When we look at the profile we can see all our functions accounted for (pprof) top10 Showing nodes accounting for 2010ms, 100% of 2010ms total Showing top 10 nodes out of 12 flat flat% sum% cum cum% 980ms 48.76% 48.76% 980ms 48.76% main.calcFib_I 910ms 45.27% 94.03% 910ms 45.27% main.calcFib_II 120ms 5.97% 100% 120ms 5.97% runtime.newstack 0 0% 100% 980ms 48.76% main.doWithLabels 0 0% 100% 980ms 48.76% main.doWithLabels.func1 0 0% 100% 980ms 48.76% main.iteratedFib_I 0 0% 100% 910ms 45.27% main.iteratedFib_II 0 0% 100% 980ms 48.76% main.main 0 0% 100% 980ms 48.76% main.rootFunc 0 0% 100% 980ms 48.76% runtime.main In the example above we can see that our `calcFib_I` and `calcFib_II` account for pretty much all of the profile time. (Please note that `calcFib_II` is the version which is called from a separate goroutine) If we then look at just the `foo` tag we see (pprof) tagfocus=foo (pprof) top10 Active filters: tagfocus=foo Showing nodes accounting for 0.91s, 45.27% of 2.01s total Showing top 10 nodes out of 12 flat flat% sum% cum cum% 0.51s 25.37% 25.37% 0.51s 25.37% main.calcFib_I 0.39s 19.40% 44.78% 0.39s 19.40% main.calcFib_II 0.01s 0.5% 45.27% 0.01s 0.5% runtime.newstack 0 0% 45.27% 0.51s 25.37% main.doWithLabels 0 0% 45.27% 0.51s 25.37% main.doWithLabels.func1 0 0% 45.27% 0.51s 25.37% main.iteratedFib_I 0 0% 45.27% 0.39s 19.40% main.iteratedFib_II 0 0% 45.27% 0.51s 25.37% main.main 0 0% 45.27% 0.51s 25.37% main.rootFunc 0 0% 45.27% 0.51s 25.37% runtime.main looking at `bar` tag we see (pprof) tagfocus=bar (pprof) top10 Active filters: tagfocus=bar Showing nodes accounting for 970ms, 48.26% of 2010ms total flat flat% sum% cum cum% 520ms 25.87% 25.87% 520ms 25.87% main.calcFib_II 450ms 22.39% 48.26% 450ms 22.39% main.calcFib_I 0 0% 48.26% 450ms 22.39% main.doWithLabels 0 0% 48.26% 450ms 22.39% main.doWithLabels.func1 0 0% 48.26% 450ms 22.39% main.iteratedFib_I 0 0% 48.26% 520ms 25.87% main.iteratedFib_II 0 0% 48.26% 450ms 22.39% main.main 0 0% 48.26% 450ms 22.39% main.rootFunc 0 0% 48.26% 450ms 22.39% runtime.main 0 0% 48.26% 450ms 22.39% runtime/pprof.Do So it behaves in a sensible way. In particular we get to see function calls in spawned goroutines grouped under the profiling labels attached to the goroutines that created them. So all is well, I think. On Thursday, 3 January 2019 16:09:09 UTC+1, Francis wrote: > > Please note the original example won't run properly. `pprof.Labels()` > takes an even number of strings, since it's setting key/value pairs. > > func handleFoo(ctx context.Context) { > labels := pprof.Labels("foo", "foo") > pprof.Do(ctx, labels, func(ctx context.Context) { > go interestingFunc() > }) > } > > Works better. > > On Thursday, 3 January 2019 15:22:06 UTC+1, Francis wrote: >> >> >> I would like to use pprof labels, but it seems like I would need to pass >> a `context.Context` into all new goroutines and re-call pprof.Do at each of >> these stages. Am I mistaken here. >> >> specifically if we had >> >> unc handleFoo(ctx context.Context) { >> pprof.Do(ctx, pprof.Labels("foo"), func(ctx context.Context) { >> go interestingFunc() >> }) >> } >> >> My understanding is that `interestingFunc` would not be profiled with the >> ‘foo’ label. >> If this is correct it presents a problem that I can’t group profiles >> under a label when I don’t control the spawning of goroutines myself. >> i.e. if a library spawns goroutines the work they do will simply appear >> under the global profile >> > -- You received this message because you are subscribed to the Google Groups "golang-nuts" group. To unsubscribe from this group and stop receiving emails from it, send an email to golang-nuts+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/d/optout.
