Is your feature request related to a problem? Please describe.
My data nodes are running Centos 7. Though that operating system is not supported by mmlspark, here is what I did to make the examples for 1.0.0-rc1 run.
Describe the solution you'd like
Download the source code for gcc-5.5.0 and openmpi-1.10.7 onto a data node.
Build and install the libraries. It will likely be necessary to install development tools to do so.
Copy the binaries to all data nodes, making sure not to overwrite any versions that the operating system is currently using. It is not necessary to cause the operating system to use the new libstdc++.so, etc.
In spark-defaults.conf add the line
spark.executor.extraLibraryPath /usr/local/lib64:/usr/lib64/openmpi/
or wherever the libraries were actually copied to.
Additional context
Use caution in deploying the gcc-5.5.0 libraries, as overwriting libraries currently in use by the operating system may make it unbootable.
Is your feature request related to a problem? Please describe.
My data nodes are running Centos 7. Though that operating system is not supported by mmlspark, here is what I did to make the examples for 1.0.0-rc1 run.
Describe the solution you'd like
spark.executor.extraLibraryPath /usr/local/lib64:/usr/lib64/openmpi/
or wherever the libraries were actually copied to.Additional context
Use caution in deploying the gcc-5.5.0 libraries, as overwriting libraries currently in use by the operating system may make it unbootable.