gopprofheap-profiling

Interpreting pprof heap diagrams


When I profile the heap in go with pprof I get the following:

enter image description here

However, I'm not clear on how to interpret that visualization. In particular:

"The memory next to the arrows means _____ and the memory inside of a box means ______. So when a box has multiple arrows from it, it means _____, and when it has multiple arrows to it, it means _____".


Solution

    1. the perceived cumulative memory consumption of this function including all of its subfunctions
    2. the perceived memory consumption of this function excluding all of its subfunctions
    3. that it calls multiple subfunctions
    4. that multiple functions call it

    No difference between the meaning of the arrows between the heap and the CPU profiling mode.

    You may be interested in the output of the top pprof command, which can give you results in the following form:

    9701.61kB of 9701.61kB total (  100%) Dropped 112 nodes (cum <=
    48.51kB) Showing top 10 nodes out of 29 (cum >= 3611.54kB)
          flat  flat%   sum%        cum   cum%
     4549.72kB 46.90% 46.90%  4549.72kB 46.90%  mystery.function
     2587.52kB 26.67% 73.57%  2587.52kB 26.67%  reflect.unsafe_NewArray
     1024.02kB 10.56% 84.12%  1024.02kB 10.56%  encoding/xml.copyValue
         514kB  5.30% 89.42%      514kB  5.30%  compress/gzip.NewReader
         514kB  5.30% 94.72%      514kB  5.30%  net/http.(*Transport).dialConn
      512.34kB  5.28%   100%   512.34kB  5.28%  runtime.makeslice
             0     0%   100%      514kB  5.30%  bytes.(*Buffer).ReadFrom
             0     0%   100%  3611.54kB 37.23%  encoding/xml.(*Decoder).Decode
             0     0%   100%  3611.54kB 37.23%  encoding/xml.(*Decoder).DecodeElement
             0     0%   100%  3611.54kB 37.23%  encoding/xml.(*Decoder).unmarshal
    

    And here are (IMHO) the two best references for Go heap profiling with pprof:

    1. https://google-perftools.googlecode.com/svn/trunk/doc/heapprofile.html
    2. https://software.intel.com/en-us/blogs/2014/05/10/debugging-performance-issues-in-go-programs