Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[ci] dowgrade gcc version for Linux on Azure #1718

Merged
merged 12 commits into from
Oct 3, 2018
Merged

[ci] dowgrade gcc version for Linux on Azure #1718

merged 12 commits into from
Oct 3, 2018

Conversation

StrikerRUS
Copy link
Collaborator

@StrikerRUS StrikerRUS commented Sep 30, 2018

Refer to #1708.

@StrikerRUS
Copy link
Collaborator Author

StrikerRUS commented Sep 30, 2018

GPU test fails. I suppose that the reason is in these lines:

2018-09-30T17:18:46.9243394Z Preparing to unpack .../ocl-icd-libopencl1_2.2.8-1_amd64.deb ...
2018-09-30T17:18:46.9646898Z Unpacking ocl-icd-libopencl1:amd64 (2.2.8-1) ...
2018-09-30T17:18:47.1272916Z Selecting previously unselected package opencl-headers.
2018-09-30T17:18:47.1380157Z Preparing to unpack .../opencl-headers_2.0~svn32091-2_all.deb ...
2018-09-30T17:18:47.1548332Z Unpacking opencl-headers (2.0~svn32091-2) ...
2018-09-30T17:18:47.2831063Z Selecting previously unselected package ocl-icd-opencl-dev:amd64.
2018-09-30T17:18:47.2923796Z Preparing to unpack .../ocl-icd-opencl-dev_2.2.8-1_amd64.deb ...
2018-09-30T17:18:47.3099881Z Unpacking ocl-icd-opencl-dev:amd64 (2.2.8-1) ...
2018-09-30T17:18:47.3866224Z Processing triggers for libc-bin (2.23-0ubuntu10) ...
2018-09-30T17:18:47.5810345Z Setting up ocl-icd-libopencl1:amd64 (2.2.8-1) ...
2018-09-30T17:18:47.6531984Z Setting up opencl-headers (2.0~svn32091-2) ...
2018-09-30T17:18:47.7040672Z Setting up ocl-icd-opencl-dev:amd64 (2.2.8-1) ...
---> 2018-09-30T17:18:47.7545016Z Processing triggers for libc-bin (2.23-0ubuntu10) ... <---

While installing OpenCL libraries libc is being updated and becomes incompatible with g++-4.8. It's my assumption.

@guolinke Since we don't provide any artifacts from GPU test publicly, what do you think about the following?

elif [[ $AZURE == "true" ]] && [[ $COMPILER == "gcc" ]] && [[ $TASK != "GPU" ]]; then
  <downgrade gcc>

@guolinke
Copy link
Collaborator

guolinke commented Oct 1, 2018

@StrikerRUS yeah, that is okay

.ci/setup.sh Outdated
if [[ $AZURE == "true" ]] && [[ $COMPILER == "clang" ]]; then
update-alternatives --install /usr/bin/clang++ clang++ /usr/bin/clang++-6.0 100
update-alternatives --install /usr/bin/clang clang /usr/bin/clang-6.0 100
if [[ $AZURE == "true" ]]; then
sudo apt-get update
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

may be we should update for travis as well, since line 48 will install ocl-icd-opencl-dev

Copy link
Collaborator Author

@StrikerRUS StrikerRUS Oct 1, 2018

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I tested without sudo apt-get update at Travis - worked well. But if it doesn't significantly enlarge the test time, lets leave this.

@guolinke
Copy link
Collaborator

guolinke commented Oct 1, 2018

@StrikerRUS
It seems the build binaries still require glibc 2.23:

guoke@test-vm3:~$ ldd lib_lightgbm.so
./lib_lightgbm.so: /lib/x86_64-linux-gnu/libm.so.6: version `GLIBC_2.23' not found (required by ./lib_lightgbm.so)
        linux-vdso.so.1 =>  (0x00007ffc472ae000)
        libstdc++.so.6 => /usr/lib/x86_64-linux-gnu/libstdc++.so.6 (0x00007f8ecbcc6000)
        libm.so.6 => /lib/x86_64-linux-gnu/libm.so.6 (0x00007f8ecb9c0000)
        libgomp.so.1 => /usr/lib/x86_64-linux-gnu/libgomp.so.1 (0x00007f8ecb7b1000)
        libgcc_s.so.1 => /lib/x86_64-linux-gnu/libgcc_s.so.1 (0x00007f8ecb59b000)
        libpthread.so.0 => /lib/x86_64-linux-gnu/libpthread.so.0 (0x00007f8ecb37d000)
        libc.so.6 => /lib/x86_64-linux-gnu/libc.so.6 (0x00007f8ecafb4000)
        /lib64/ld-linux-x86-64.so.2 (0x00007f8ecc445000)

So i add some fixes .

@@ -62,6 +62,10 @@ typedef void(*AllgatherFunction)(char* input, comm_size_t input_size, const comm
#define __func__ __FUNCTION__
#endif

#ifdef GCC
__asm__(".symver lgamma, lgamma@GLIBC_2.2.5");
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

maybe a better solution is implementing a lgamma by ourself.

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Hmm, why is this symver needed?

Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

lgamma will introduce the glibc2.23 dependency.

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This sounds a bit odd. I've checked the glibc changelog and it seems lgamma have been there for quite a while.

Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think lgamma have a new implementation in glibc2.23. and result in this new dependency.

CMakeLists.txt Outdated
@@ -17,6 +17,10 @@ if(APPLE)
OPTION(APPLE_OUTPUT_DYLIB "Output dylib shared library" OFF)
endif()

if((CMAKE_CXX_COMPILER_ID STREQUAL "GNU") AND (UNIX) AND (NOT APPLE))
ADD_DEFINITIONS(-DGCC)
endif()
Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Maybe it's better to move this somewhere lower to other definitions?

Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

sorry, I didn't understand it, can you describe more details ?

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@guolinke
Copy link
Collaborator

guolinke commented Oct 1, 2018

@StrikerRUS what do you think about the current fix for lgamma ?

@StrikerRUS
Copy link
Collaborator Author

@guolinke To be honest, I'm not familiar with SymbolVersioning, but I think it's rather hacky solution and intuitively I can't say that I like it. I'm afraid that it will cause some troubles.

I'd better prefer this your proposal:

maybe a better solution is implementing a lgamma by ourself.

Can we just copy-paste this function from the sources?

@guolinke
Copy link
Collaborator

guolinke commented Oct 2, 2018

@StrikerRUS I just notice we don't need the lgamma. The new fix is pushed.

guoke@test-vm3:~$ ldd lib_lightgbm.so
        linux-vdso.so.1 =>  (0x00007ffdfc5e5000)
        libstdc++.so.6 => /usr/lib/x86_64-linux-gnu/libstdc++.so.6 (0x00007f1bc6a8e000)
        libm.so.6 => /lib/x86_64-linux-gnu/libm.so.6 (0x00007f1bc6788000)
        libgomp.so.1 => /usr/lib/x86_64-linux-gnu/libgomp.so.1 (0x00007f1bc6579000)
        libgcc_s.so.1 => /lib/x86_64-linux-gnu/libgcc_s.so.1 (0x00007f1bc6363000)
        libpthread.so.0 => /lib/x86_64-linux-gnu/libpthread.so.0 (0x00007f1bc6145000)
        libc.so.6 => /lib/x86_64-linux-gnu/libc.so.6 (0x00007f1bc5d7c000)
        /lib64/ld-linux-x86-64.so.2 (0x00007f1bc720d000)

@superbobry
Copy link

What is the glibc version used by GCC 4.8? Could you mention it in the PR description?

@StrikerRUS
Copy link
Collaborator Author

@guolinke

I just notice we don't need the lgamma

Wow, it's great! This fix is for sure much better than symbol versioning. 😃

@StrikerRUS
Copy link
Collaborator Author

@guolinke I've added ... || exit -1 to prevent silent usage of inappropriate gcc.

@guolinke
Copy link
Collaborator

guolinke commented Oct 2, 2018

@superbobry

guoke@rdnn-gpu-nv-01:~$ objdump -T lib_lightgbm.so | fgrep GLIBC_
0000000000000000      DF *UND*  0000000000000000  GLIBC_2.3.4 __printf_chk
0000000000000000      DF *UND*  0000000000000000  GLIBC_2.2.5 __errno_location
0000000000000000      DO *UND*  0000000000000000  GLIBC_2.2.5 stdout
0000000000000000      DF *UND*  0000000000000000  GLIBC_2.2.5 bind
0000000000000000      DF *UND*  0000000000000000  GLIBC_2.2.5 ceil
0000000000000000      DF *UND*  0000000000000000  GLIBC_2.2.5 socket
0000000000000000      DF *UND*  0000000000000000  GLIBC_2.3.4 __vfprintf_chk
0000000000000000      DF *UND*  0000000000000000  GLIBC_2.2.5 nanosleep
0000000000000000      DF *UND*  0000000000000000  GLIBC_2.2.5 accept
0000000000000000      DF *UND*  0000000000000000  GLIBC_2.2.5 log
0000000000000000      DF *UND*  0000000000000000  GLIBC_2.2.5 recv
0000000000000000  w   DF *UND*  0000000000000000  GLIBC_2.2.5 __cxa_finalize
0000000000000000      DF *UND*  0000000000000000  GLIBC_2.2.5 strlen
0000000000000000      DF *UND*  0000000000000000  GLIBC_2.2.5 memcmp
0000000000000000      DF *UND*  0000000000000000  GLIBC_2.2.5 send
0000000000000000      DF *UND*  0000000000000000  GLIBC_2.2.5 memset
0000000000000000      DF *UND*  0000000000000000  GLIBC_2.2.5 pthread_create
0000000000000000      DF *UND*  0000000000000000  GLIBC_2.3.4 __sprintf_chk
0000000000000000      DF *UND*  0000000000000000  GLIBC_2.3.4 __snprintf_chk
0000000000000000      DF *UND*  0000000000000000  GLIBC_2.2.5 vsnprintf
0000000000000000      DF *UND*  0000000000000000  GLIBC_2.2.5 __assert_fail
0000000000000000      DF *UND*  0000000000000000  GLIBC_2.2.5 nextafter
0000000000000000  w   DF *UND*  0000000000000000  GLIBC_2.2.5 pthread_mutex_unlock
0000000000000000      DF *UND*  0000000000000000  GLIBC_2.2.5 strtod
0000000000000000      DF *UND*  0000000000000000  GLIBC_2.2.5 sqrt
0000000000000000      DF *UND*  0000000000000000  GLIBC_2.14  memcpy
0000000000000000      DF *UND*  0000000000000000  GLIBC_2.2.5 connect
0000000000000000      DF *UND*  0000000000000000  GLIBC_2.3.4 __strcpy_chk
0000000000000000      DF *UND*  0000000000000000  GLIBC_2.2.5 __cxa_atexit
0000000000000000      DF *UND*  0000000000000000  GLIBC_2.2.5 pow
0000000000000000      DF *UND*  0000000000000000  GLIBC_2.2.5 fclose
0000000000000000      DF *UND*  0000000000000000  GLIBC_2.2.5 inet_ntop
0000000000000000      DF *UND*  0000000000000000  GLIBC_2.4   __stack_chk_fail
0000000000000000      DF *UND*  0000000000000000  GLIBC_2.2.5 listen
0000000000000000      DF *UND*  0000000000000000  GLIBC_2.2.5 fflush
0000000000000000      DF *UND*  0000000000000000  GLIBC_2.2.5 fopen
0000000000000000      DF *UND*  0000000000000000  GLIBC_2.2.5 exp
0000000000000000      DO *UND*  0000000000000000  GLIBC_2.2.5 stderr
0000000000000000      DF *UND*  0000000000000000  GLIBC_2.3.4 __vsprintf_chk
0000000000000000      DF *UND*  0000000000000000  GLIBC_2.2.5 strtol
0000000000000000  w   DF *UND*  0000000000000000  GLIBC_2.2.5 pthread_mutex_lock
0000000000000000      DF *UND*  0000000000000000  GLIBC_2.2.5 fread
0000000000000000      DF *UND*  0000000000000000  GLIBC_2.2.5 log2f
0000000000000000      DF *UND*  0000000000000000  GLIBC_2.2.5 log2
0000000000000000      DF *UND*  0000000000000000  GLIBC_2.2.5 memmove
0000000000000000      DF *UND*  0000000000000000  GLIBC_2.3.4 __fprintf_chk
0000000000000000  w   DF *UND*  0000000000000000  GLIBC_2.2.5 __pthread_key_create
0000000000000000      DF *UND*  0000000000000000  GLIBC_2.3   __tls_get_addr
0000000000000000      DF *UND*  0000000000000000  GLIBC_2.3   freeifaddrs
0000000000000000      DF *UND*  0000000000000000  GLIBC_2.2.5 setsockopt
0000000000000000      DF *UND*  0000000000000000  GLIBC_2.2.5 sqrtf
0000000000000000      DF *UND*  0000000000000000  GLIBC_2.3   getifaddrs
0000000000000000      DF *UND*  0000000000000000  GLIBC_2.2.5 inet_pton
0000000000000000      DF *UND*  0000000000000000  GLIBC_2.2.5 fwrite
0000000000000000      DF *UND*  0000000000000000  GLIBC_2.2.5 close
0000000000000000      DF *UND*  0000000000000000  GLIBC_2.2.5 logf

@guolinke guolinke merged commit 52dbc57 into master Oct 3, 2018
@superbobry
Copy link

Should we expect a release following this PR?

@StrikerRUS StrikerRUS deleted the glibc branch October 3, 2018 10:08
@guolinke
Copy link
Collaborator

guolinke commented Oct 3, 2018

yeah, refer to #1727

Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

None yet

4 participants