grpc library bloat

[mcfadden8]

I have spent the last week or so attempting to build one of our production codes with our new SCR checkpoint library that makes use of the nnfdm.a library and have run in to some issues that I think could be a pain point for some of our code teams (thus far, I am still unable to successfully build). Library bloat. More specifically, because the nnfdm api is using grpc with protobuf, this is causing us to link in 57 (not an exaggeration) additional libraries (shown below).

Does anyone know of a lighter weight way to trim down the large number of libraries? With our codes that have complicated build systems, it is not a trivial task to deal handle the larger set of libraries that seem to be required.

lib64/libabsl_bad_optional_access.a lib64/libabsl_bad_variant_access.a lib64/libabsl_base.a lib64/libabsl_city.a lib64/libabsl_civil_time.a lib64/libabsl_cord.a lib64/libabsl_cord_internal.a lib64/libabsl_cordz_functions.a lib64/libabsl_cordz_handle.a lib64/libabsl_cordz_info.a lib64/libabsl_debugging_internal.a lib64/libabsl_demangle_internal.a lib64/libabsl_exponential_biased.a lib64/libabsl_graphcycles_internal.a lib64/libabsl_hash.a lib64/libabsl_hashtablez_sampler.a lib64/libabsl_int128.a lib64/libabsl_log_severity.a lib64/libabsl_low_level_hash.a lib64/libabsl_malloc_internal.a lib64/libabsl_random_distributions.a lib64/libabsl_random_internal_platform.a lib64/libabsl_random_internal_pool_urbg.a lib64/libabsl_random_internal_randen.a lib64/libabsl_random_internal_randen_hwaes.a lib64/libabsl_random_internal_randen_hwaes_impl.a lib64/libabsl_random_internal_randen_slow.a lib64/libabsl_random_internal_seed_material.a lib64/libabsl_random_seed_gen_exception.a lib64/libabsl_random_seed_sequences.a lib64/libabsl_raw_hash_set.a lib64/libabsl_raw_logging_internal.a lib64/libabsl_spinlock_wait.a lib64/libabsl_stacktrace.a lib64/libabsl_status.a lib64/libabsl_statusor.a lib64/libabsl_str_format_internal.a lib64/libabsl_strings.a lib64/libabsl_strings_internal.a lib64/libabsl_symbolize.a lib64/libabsl_synchronization.a lib64/libabsl_throw_delegate.a lib64/libabsl_time.a lib64/libabsl_time_zone.a lib64/libprotobuf.a lib/libaddress_sorting.a lib/libcares.a lib/libcrypto.a lib/libgpr.a lib/libgrpc++.a lib/libgrpc.a lib/libgrpc++_reflection.a lib/libnnfdm.a lib/libre2.a lib/libssl.a lib/libupb.a lib/libz.`

[mcfadden8]

This is likely caused by how I am building grpc (which is required since we need a later version than what is available with TOSS (if I remember correctly)).

I am hoping that I can build a much liter version of grpc since I would find it hard to believe we need everything+kitchensink

[bdevcich]

This is likely caused by how I am building grpc (which is required since we need a later version than what is available with TOSS (if I remember correctly)).

I am hoping that I can build a much liter version of grpc since I would find it hard to believe we need everything+kitchensink

Which version is TOSS using?

[mcfadden8]

I actually do not know. It is possible that it isn't even installed on rzadams. It has been so long since we last checked. @behlendorf might be able to say which version is there (if any at all).

[mcfadden8]

Here is how I currently am building gRPC and nnfdm

rm -rf  /g/g0/martymcf/libraries/nnfdm.master/toss_4_x86_64_ib_cray/grpc.v1.46.3 \
        /g/g0/martymcf/libraries/nnfdm.master/toss_4_x86_64_ib_cray/nnfdm.master \
        /g/g0/martymcf/libraries/nnfdm.master/toss_4_x86_64_ib_cray/install

git clone --recurse-submodules --branch v1.46.3 --depth 1 --shallow-submodules \
    [email protected]:grpc/grpc /g/g0/martymcf/source/nnfdm.master/grpc.v1.46.3

mkdir -p /g/g0/martymcf/libraries/nnfdm.master/toss_4_x86_64_ib_cray/grpc.v1.46.3 \
cd /g/g0/martymcf/libraries/nnfdm.master/toss_4_x86_64_ib_cray/grpc.v1.46.3 \
flux run -t 120 -N1 \
  cmake -Wno-dev \
    -DCMAKE_CXX_COMPILER=/opt/rocm-6.1.2/bin/amdclang++ \
    -DCMAKE_C_COMPILER=/opt/rocm-6.1.2/bin/amdclang \
    -DCMAKE_INSTALL_PREFIX=/g/g0/martymcf/libraries/nnfdm.master/toss_4_x86_64_ib_cray/install \
    -DgRPC_INSTALL=ON \
    -DgRPC_BUILD_TESTS=OFF \
    /g/g0/martymcf/source/nnfdm.master/grpc.v1.46.3

cd /g/g0/martymcf/libraries/nnfdm.master/toss_4_x86_64_ib_cray/grpc.v1.46.3 
flux run -t 120 -N1 make install

CMAKE_PREFIX_PATH=/g/g0/martymcf/libraries/nnfdm.master/toss_4_x86_64_ib_cray/install \
  mkdir -p /g/g0/martymcf/libraries/nnfdm.master/toss_4_x86_64_ib_cray/nnfdm.master && \
  cd /g/g0/martymcf/libraries/nnfdm.master/toss_4_x86_64_ib_cray/nnfdm.master && \
  flux run -t 120 -N1 \
    cmake -Wno-dev -DCMAKE_POSITION_INDEPENDENT_CODE=ON -DBUILD_SHARED_LIBS=Off \
          -DCMAKE_CXX_COMPILER=/opt/rocm-6.1.2/bin/amdclang++ \
          -DCMAKE_C_COMPILER=/opt/rocm-6.1.2/bin/amdclang \
          -DCMAKE_INSTALL_PREFIX=/g/g0/martymcf/libraries/nnfdm.master/toss_4_x86_64_ib_cray/install \
          /g/g0/martymcf/source/nnfdm.master/nnfdm.master/daemons/compute/lib-cpp

cd /g/g0/martymcf/libraries/nnfdm.master/toss_4_x86_64_ib_cray/nnfdm.master 
flux run -t 120 -N1 make install

[mcfadden8]

Actually, I think that gRPC for C++ requires manual installation (at least according to https://grpc.io/blog/installation/)