torchrec
torchrec copied to clipboard
pytorch
Fix dynamic embedding bugs that may lead to segmentation faults
Hi TorchRec Team,
We've recently been trying to use the dynamic embedding feature in torchrec contrib, but we have encountered a few challenges. The process may result in a segmentation fault. After debugging, we've identified two potential problems.
1） fetch_notifications_ is not thread-safe. 2）Tensor data might be recycled during pushing operations.
For the first issue, the fetch_notifications_ variable is accessed in both the pull and sync fetch functions. As these functions are called in different threads, there might be a thread safety issue.
For the second issue, the tensor generated by concat in the push function is a temporary variable. The subsequent call to io.push is asynchronous and does not copy the data. Therefore, the data might have been recycled by the time it is executed asynchronously, leading to access to invalid data.
In light of these findings, we have undertaken some corrective measures to address these issues. We appreciate your attention to this Pull Request and eagerly await your valuable feedback.
Thank you for your time and consideration.
Hi @sherlockkenan!
Thank you for your pull request and welcome to our community.
Action Required
In order to merge any pull request (code, docs, etc.), we require contributors to sign our Contributor License Agreement, and we don't seem to have one on file for you.
Process
In order for us to review and merge your suggested changes, please sign at https://code.facebook.com/cla. If you are contributing on behalf of someone else (eg your employer), the individual CLA may not be sufficient and your employer may need to sign the corporate CLA.
Once the CLA is signed, our tooling will perform checks and validations. Afterwards, the pull request will be tagged with CLA signed. The tagging process may take up to 1 hour after signing. Please give it that time before contacting us about it.
If you have received this in error or have any questions, please contact us at [email protected]. Thanks!
Thank you for signing our Contributor License Agreement. We can now accept your code for this (and any) Meta Open Source project. Thanks!
Hi, I don't have expertise on dynamic embeddings, so probably will just cc @colin2328
also probably make an issue as well, since it is easier to track there
sorry for the late response. I have created an issue here: https://github.com/pytorch/torchrec/issues/2006. I would appreciate any further suggestions regarding the pull request.
@PaulZhang12 has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator.
Please address CI failures and then you are good
thanks! sorry for the failures and I've fixed them and pushed the changes to the merge request. However, it seems that the workflow requires approval again.