prov/util: Switch ep_list_lock to always be mutex #9377
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
|
I do not think this is the right long term solution, but it gives us time to come up with a better option (without fully reverting). |
|
I agree that we don't want to revert completely. We will also need to update the other places where this change was made. Counter and EQ? (I don't remember). FWIW, I think apps using msg ep's may be more impacted by the threading options than rdm ep's. That's where my concern is coming from. |
Switch ep_list_lock to always be a mutex lock until we can agree on a better long term decision. The issue is that some endpoints want the ability to not require FI_THREAD_SAFE at the domain level, but they want to be able to control the endpoint in a seperate thread. The ep_list_lock is used for both the data path, and the control path. In order to adhere to the current api, and not break providers that follow the current api, we need to revert this lock to a mutex until we can get a optimal solution for everyone. Signed-off-by: Seth Zegelstein <szegel@amazon.com>
a-szegel
force-pushed
the
make_utilcq_always_use_mutex
branch
from
bf28890
to
40ffd2b
Compare
|
Done. I don't want to trade use cases. I want to get the optimal approach for both use cases. |
|
I agree. However I think doing so may require changing the behavior outlined in the man pages. That's doable for version 2, but may be difficult in v1. |
a-szegel
added a commit
to a-szegel/libfabric
that referenced
this pull request
Oct 23, 2023
I mistakenly reverted the wrong lock on PR#9377, and caused a performance regression. This change makes the ep->lock be domain dependent, and properly revers util_cntr ep_list_lock to always be mutex. This restores most of the performance regression. Signed-off-by: Seth Zegelstein <szegel@amazon.com>
a-szegel
added a commit
to a-szegel/libfabric
that referenced
this pull request
Oct 23, 2023
I mistakenly reverted ep->lock to always be an ofi_mutex. Restore ep->lock to depend on domain threading model Signed-off-by: Seth Zegelstein <szegel@amazon.com>
a-szegel
added a commit
to a-szegel/libfabric
that referenced
this pull request
Oct 25, 2023
I mistakenly reverted ep->lock to always be an ofi_mutex. Restore ep->lock to depend on domain threading model Signed-off-by: Seth Zegelstein <szegel@amazon.com>
j-xiong
pushed a commit
that referenced
this pull request
Oct 26, 2023
I mistakenly reverted ep->lock to always be an ofi_mutex. Restore ep->lock to depend on domain threading model Signed-off-by: Seth Zegelstein <szegel@amazon.com>
Juee14Desai
pushed a commit
to Juee14Desai/libfabric
that referenced
this pull request
Jan 31, 2024
I mistakenly reverted ep->lock to always be an ofi_mutex. Restore ep->lock to depend on domain threading model Signed-off-by: Seth Zegelstein <szegel@amazon.com>
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.
Switch ep_list_lock to always be a mutex lock until we can agree on a better long term decision. The issue is that some endpoints want the ability to not require FI_THREAD_SAFE at the domain level, but they want to be able to control the endpoint in a seperate thread. The ep_list_lock is used for both the data path, and the control path. In order to adhere to the current api, and not break providers that follow the current api, we need to revert this lock to a mutex until we can get a optimal solution for everyone.