Integrate valgrind in CircleCI build target

[fuchsto]

Valgrind should be added in at least a minimal CI target to check for memory leaks and consistency defects.

[devreal]

I would prefer someone already familiar with CircleCI to do the integration as I cannot give this a high priority right now. It will also be much quicker to do for the maintainers of the CI scripts. Maybe @fmoessbauer can take over this task?

[fuchsto]

@fmoessbauer is sick for the next days, actually it's the same situation here, it's not like CI is a core task for us either.
Also, I would prefer if the whole team, or at least more than 1-2 members, understand CI, otherwise the team in Munich degenerates to second level support. We have been new to CI two weeks ago as well.

And valgrind integration is not a mission critical feature of course, so I didn't expect a quick implementation of this issue request, we just should have it in our tasks.

[fuchsto]

@devreal But, speaking of it: we should provide a documentation of our CI procedure to make contributions easier for other team members.

Documentation of CI is definitely a task for @fmoessbauer and me (@fuchsto).

See support issue #76

Once documentation is available, we should discuss the current CI procedure and define all required integration test scenarios in a brief conference call.

[devreal]

I agree that there should be more than one person understanding the CI. I had a look at some of the scripts yesterday and it was not clear to me where to start with the intergration. Once we have documentation from #76 I'm happy to look into that.

[fuchsto]

Perfect, thanks!

Yes, the CI scripts need some cleanup, shot myself in the foot quite frequently myself.
We will consolidate deployment first to make things less cryptic.

[devreal]

I have been playing around with this today, mainly trying to get the CI and docker working locally (I still cannot build a docker container locally due to network issues). My first shot is at branch sup-75-valgrindci, please have a look at it. I am created a new docker configuration that installs valgrind and sets EXEC_WRAP, which is then used as a prefix to the test executable in dash-test.sh. Comments are welcome.

One thing that is still missing: I think it only makes sense to run valgrind under the debug configuration so there is no need to also run the other configurations. I haven't figured out how to pass this information to the scripts so far.

[fmoessbauer]

Nice Job, looks great! I implemented the described behavior by just skipping the main loop if container is openmpi2_vg and target is not Debug. I also extended the documentation according to this.

I also added the container to dockerhub, as otherwise it cannot be pulled. Have to re-run CI after the build of the container has finished.

[fuchsto]

@devreal Awesome!

@fmoessbauer Remember to use the valgrind MPI wrappers, otherwise we will get lots of false positives from valgrind in the MPI runtime.

[fmoessbauer]

Don't be happy to early. There are still some todos.

• Currently the command is executed the following way (from CI output). Is that correct?
  mpirun --allow-run-as-root --map-by core -n 1 valgrind --track-origins=yes --leak-check=full /opt/dash/build-ci/20161111-180312/Debug/bin/dash/test/mpi/dash-test-mpi
• We need a detection of leak messages in the logs (can be easily added in dash-ci.sh)
• guarantee that not to much output is produces. See this build for details.