CPU statistics using eBPF

@thiagoftsm commented on Mon Mar 09 2020

Feature idea summary

During the spike #7885 , it was identified that we can bring CPU monitoring using perf events like the command perf does. And considering what we have as output from this command, it is possible to say that we can bring a general CPU statistic using perf events.

Expected behavior

To have new charts that will demonstrate the following information per PID:

       1920.159821 task-clock                #    0.991 CPUs utilized          
                13 context-switches          #    0.007 K/sec                  
                 0 CPU-migrations            #    0.000 K/sec                  
               258 page-faults               #    0.134 K/sec                  
     5,649,595,479 cycles                    #    2.942 GHz                     [83.43%]
     1,808,339,931 stalled-cycles-frontend   #   32.01% frontend cycles idle    [83.54%]
     1,171,884,577 stalled-cycles-backend    #   20.74% backend  cycles idle    [66.77%]
     8,625,207,199 instructions              #    1.53  insns per cycle        
                                             #    0.21  stalled cycles per insn [83.51%]
     1,488,797,176 branches                  #  775.351 M/sec                   [82.58%]
        53,395,139 branch-misses             #    3.59% of all branches         [83.78%]

Ths example was extracted from http://brendangregg.com/perf.html#CPUstatistics

Start by reviewing spike #7885 and the linked perf CPU statistics reference to understand the proposed event set. The issue names no files or tests, so first locate the relevant eBPF collector entry points. Done means adding per-PID charts for the listed CPU, scheduling, memory, cycle, instruction, branch, and miss statistics.