Umpire shares its Uberenv workflow with other projects. The documentation is therefore shared.
This page will provides some Umpire specific examples to illustrate the workflow described in the documentation.
Before to start¶
First of all, it is worth noting that Umpire does not have dependencies, except for CMake, which is most of the time installed externally.
That does not make the workflow useless: Uberenv will drive Spack which will generate a host-config file with the toolchain (including cuda if activated) and the options or variants pre-configured.
Machine specific configuration¶
$ ls -c1 scripts/uberenv/spack_configs blueos_3_ppc64le_ib darwin toss_3_x86_64_ib blueos_3_ppc64le_ib_p9 config.yaml
Umpire has been configured for
toss_3_x86_64_ib and other systems.
$ ls -c1 .gitlab/*jobs.yml .gitlab/lassen-jobs.yml .gitlab/ruby-jobs.yml
CI contains jobs for ruby.
$ git grep -h "SPEC" .gitlab/ruby-jobs.yml | grep "gcc" SPEC: "%email@example.com" SPEC: "%firstname.lastname@example.org" SPEC: "%email@example.com" SPEC: "%firstname.lastname@example.org" SPEC: "%email@example.com"
We now have a list of the specs vetted on
In practice, one should check if the job is not allowed to fail, or even deactivated.
In Umpire, the Spack configuration for MacOS contains the default compilers depending on the OS version (compilers.yaml), and a commented section to illustrate how to add CMake as an external package. You may install CMake with homebrew, for example.
Using Uberenv to generate the host-config file¶
We have seen that we can safely use firstname.lastname@example.org on ruby. Let us ask for the default configuration first, and then produce static libs, have OpenMP support and run the benchmarks:
$ python scripts/uberenv/uberenv.py --spec="%email@example.com" $ python scripts/uberenv/uberenv.py --spec="%firstname.lastname@example.org~shared+openmp tests=benchmarks"
Each will generate a CMake cache file, e.g.:
Using host-config files to build Umpire¶
$ mkdir build && cd build $ cmake -C <path_to>/<host-config>.cmake .. $ cmake --build -j . $ ctest --output-on-failure -T test
It is also possible to use this configuration with the CI script outside of CI:
$ HOST_CONFIG=<path_to>/<host-config>.cmake scripts/gitlab/build_and_test.sh
Using Uberenv to configure and run Leak Sanitizer¶
During development, it may be beneficial to regularly check for memory leaks. This will help avoid the possibility of having many memory leaks showing up all at once during the CI tests later on. The Leak Sanitizer can easily be configured from the root directory with:
$ srun -ppdebug -N1 --exclusive python scripts/uberenv/uberenv.py --spec="%email@example.com cxxflags=-fsanitize=address" $ cd build $ cmake -C <path_to>/firstname.lastname@example.org .. $ cmake --build -j $ ASAN_OPTIONS=detect_leaks=1 make test
The host config file (i.e.,
hc-ruby-...cmake) can be reused in order to rebuild with the same configuration if needed.
This will configure a build with Clang 9.0.0 and the Leak Sanitizer. If there is a leak in one of the tests, it can be useful to gather more information about what happened and more details about where it happened. One way to do this is to run:
$ ASAN_OPTIONS=detect_leaks=1 ctest -T test --output-on-failure
Additionally, the Leak Sanitizer can be run on one specific test (in this example, the “replay” tests) with:
$ ASAN_OPTIONS=detect_leaks=1 ctest -T test -R replay --output-on-failure
Depending on the output given when running the test with the Leak Sanitizer, it may be useful to use
addr2line -e <./path_to/executable> <address_of_leak> to see the exact line the output is referring to.