Change linking order to avoid using gloo in pytorch dynamic libraries #3750
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Signed-off-by: cheng.chen <c392410997@gmail.com>
Unit Test Results (with flaky tests) 1 382 files + 42 1 382 suites +42 14h 1m 40s ⏱️ + 36m 39s For more details on these failures, see this check. Results for commit 9f66283. ± Comparison against base commit 44f9357. ♻️ This comment has been updated with latest results. |
maxhgerlach
approved these changes
Oct 18, 2022
|
Here are the tests, all green (I had to rerun the failed "CI (Results)" workflow). |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Signed-off-by: cheng.chen c392410997@gmail.com
Description
{Pytorch_LIBRARIES} contains libtorch_cpu.so, which also includes gloo-related symbols, if {Pytorch_LIBRARIES} comes before gloo/compatible_gloo, then some of gloo functions (under gloo:rendezvous namespace) would dynamic link to libtorch_cpu.so instead of static link to gloo/compatible_gloo in horovod. This would create incompatible related errors if pytorch and horovod using different gloo versions.
In my case, I use different version of gloo in pytorch and horovod, horovod generates
[1]<stderr>:*** Error in python3: malloc(): memory corruption: 0x000014578c00b6a0 ***, and I found out error generated because https://github.com/horovod/horovod/blob/master/horovod/common/gloo/gloo_context.cc#L85 links to libtorch_cpu.so, so made error at runtime.So make ${Pytorch_LIBRARIES} after gloo/compatible_gloo can make sure mpi_lib_v2.cpython-36m-x86_64-linux-gnu.so static link to the gloo/compatible_gloo of third-party in horovod.
Review process to land