[migrated from https://github.com/dmlc/xgboost/issues/3511]
Compile environment
gcc version: 6.1.0
I bulid xgboost-spark.
Can I run on gcc 4.4.7 version cluster? Thank you!
java.lang.UnsatisfiedLinkError: /usr/lib64/libstdc++.so.6: version `CXXABI_1.3.7' not found
XGBoost extensively uses features of C++11, which is fully supported by GCC 4.8.1 and higher.
You have a few options:
- Upgrade GCC on cluster. This may not be possible if you don’t have admin privileges.
- Statically link stdlibc++. Edit https://github.com/dmlc/xgboost/blob/e2f09db77a4ea1f759f9be585ab01a320d933900/CMakeLists.txt#L221 to
target_link_libraries(xgboost4j
${LINK_LIBRARIES}
${JAVA_JVM_LIBRARY} -static-libgcc -static-libstdc++)
- Copy libstdc++.so.6 from your machine to the cluster and edit
PATH
Let me know which method works the best.
Thank you. According to your method, you have successfully run xgboost v0.7 on spark 2.1.0 version. However, due to the problem of cluster python 2.6 version, we need to update /usr/lib64/libgomp.so.1.0.0.
Can you try linking libgomp statically as well?
Or you could try deploying libgomp along with the JAR.