DPDK Build Issue about omnireduce HOT 5 CLOSED

I can go forward by adding a SKIP_DPDK, but then it fails at building PyTorch and says it cannot find Gloo.

AT_INSTALL_INCLUDE_DIR include/ATen/core
core header install: /root/tmp/omnireduce/omnireduce-DPDK/pytorch/build/aten/src/ATen/core/TensorBody.h
-- /usr/bin/c++ /root/tmp/omnireduce/omnireduce-DPDK/pytorch/torch/abi-check.cpp -o /root/tmp/omnireduce/omnireduce-DPDK/pytorch/build/abi-check
-- Determined _GLIBCXX_USE_CXX11_ABI=1
CMake Error at /usr/share/cmake/Modules/FindPackageHandleStandardArgs.cmake:230 (message):
Could NOT find Gloo (missing: Gloo_INCLUDE_DIR Gloo_LIBRARY)
Call Stack (most recent call first):
/usr/share/cmake/Modules/FindPackageHandleStandardArgs.cmake:594 (_FPHSA_FAILURE_MESSAGE)
cmake/Modules/FindGloo.cmake:44 (find_package_handle_standard_args)
torch/lib/c10d/CMakeLists.txt:76 (find_package)

Although Gloo builds and installs correctly before this

-- Configuring done
-- Generating done
-- Build files have been written to: /root/tmp/omnireduce/omnireduce-DPDK/gloo/build

• make install
  Consolidate compiler generated dependencies of target gloo
  [100%] Built target gloo
  Install the project...

from omnireduce.

Are you building DPDK inside docker container?
We install DPDK in the host machine and didn't see this problem before.
Your NIC supports RoCE so you may try the RDMA version instead.

from omnireduce.

Yes I am building inside the docker. That should not be the issue here as the newer DPDK version does compile fine, might be some incompatibility with the newer OS or arch. But anyways even skipping the DPDK, how to solve the Gloo issue?

For my particular use-case, I need to run PyTorch/Gloo/DAIET/DPDK so have to make this one work. Right now the only blocking issue is that pytorch make cannot fine Gloo Could NOT find Gloo (missing: Gloo_INCLUDE_DIR Gloo_LIBRARY)

from omnireduce.

If you were following the doc, you should have a conda environment. Check if you do have gloo installed in your ${CONDA_PREFIX}
If yes, try passing Gloo_INCLUDE_DIR=${CONDA_PREFIX}/include and Gloo_LIBRARY=${CONDA_PREFIX}/lib when building PyTorch as the error message says those are missing.

from omnireduce.

Fixed! I wasn't sure what to pass as the paths for include and lib directories, this worked.

from omnireduce.