Skip to Content

Profiling

During development

The recommended way to profile ingraind binaries is using perf. The default --release builds are not stripped, therefore it should be easy to analyse the results.

sudo perf record ./target/release/ingraind
sudo perf report

Docker

If running in a container, the recommended way is to start ingraind, then attach perf, and look the results off the box.

sudo docker run -d --name ingraind -e OPTION=value [...] --pid=host --net=host --privileged ingraind
sudo perf record -a -p `pgrep ingraind`

Looking at the results in this scenario is a bit tricky, because we will need to tell perf where the binaries are located to resolve the symbols.

This can be done by starting an ingraind container on the box that we’re using for analysis (if different from the system where we collected the data), then using the merged overlay filesystem as a base for symbol resolution. Note, you will need to use the same container version, as the symbols will change between builds.

This can be easily done like so:

export VERSION=sha256_of_profiled_container
docker run -d --rm --name ingraind -it ingraind:$VERSION /bin/sh

export CONTAINER_HASH=$(docker inspect ingraind |grep MergedDir |sed 's;.*: "\(.*\)",;\1;')
sudo perf report -f -i perf.data_docker --symfs /var/lib/docker/overlay2/$CONTAINER_HASH/merged

After you’re done, you can shut down the container.

docker stop ingraind

Stat

To get detailed stats about the execution of ingraind, you can perf stat like so:

$ perf stat -a -p `pgrep ingraind`

Performance counter stats for process id '28037':

   3593.405601      cpu-clock (msec)          #    0.114 CPUs utilized
       147,307      context-switches          #    0.041 M/sec
         1,611      cpu-migrations            #    0.448 K/sec
            60      page-faults               #    0.017 K/sec
 6,076,570,590      cycles                    #    1.691 GHz
 4,317,115,815      instructions              #    0.71  insn per cycle
   788,078,499      branches                  #  219.312 M/sec
    29,054,771      branch-misses             #    3.69% of all branches

  31.398821987 seconds time elapsed
Back to top