Register UCX close callback #5474
Merged
Conversation
This file contains 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
Register a close callback function with UCX to prevent writing when the endpoint has already closed. This prevents errors often raised when a remote process closes too quickly before the local process is able to send the close message. Closes rapidsai/dask-cuda#713
Ensure the scheduler has some time to close BatchedSend before closing local BatchedSend instances in Worker.
quasiben
reviewed
Nov 1, 2021
| @@ -1510,6 +1510,13 @@ async def close( | |||
|
|
|||
| self.stop_services() | |||
|
|
|||
| # Give some time for a UCX scheduler to complete closing endpoints | |||
There was a problem hiding this comment.
Thanks for the detailed comments here
|
Thanks for continuing to push on EP handling and worker closing issues with UCX. The failures here are unrelated to the PR. Merging in |
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Outdated suggestions cannot be applied.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.
With UCX, an endpoint is considered invalid to write immediately after it's closed or an error is raised due to the remote endpoint having closed. However, some messages may have already arrived and enqueued at the UCX endpoint but not having been read by the application. We must ensure that an endpoint is not valid for writing anymore immediately after an error has happened, but reading can still be done while there are enqueued, unread messages. By registering a close callback we can achieve that to prevent unclean closing of Distributed comms.
This change depends on rapidsai/ucx-py#795 for clean closing, but checks ensure using an older UCX-Py build without that feature still works.
Closes rapidsai/dask-cuda#713 .