Fix Race condition in --async-offload that can cause corruption #10501
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
rattus128
force-pushed
the
async-offload-streams
branch
from
e61b839 to
a50c8c2
Compare
|
This one would do well with an integration test, is there a way to integ test with --async-offload set? |
|
Nice! Talked with comfy and this will have to go in after #10498 is merged, so this PR will need to be rebased on that one. |
|
Can you rebase this on the latest? |
Make this a reusable function.
This sync is nessacary as pytorch will queue cuda async frees on the same stream as created to tensor. In the case of async offload, this will be on the offload stream. Weights and biases can go out of scope in python which then triggers the pytorch garbage collector to queue the free operation on the offload stream possible before the compute stream has used the weight. This causes a use after free on weight data leading to total corruption of some workflows. So sync the offload stream with the compute stream after the weight has been used so the free has to wait for the weight to be used. The cast_bias_weight is extended in a backwards compatible way with the new behaviour opt-in on a defaulted parameter. This handles custom node packs calling cast_bias_weight and defeatures async-offload for them (as they do not handle the race). The pattern is now: cast_bias_weight(... , offloadable=True) #This might be offloaded thing(weight, bias, ...) uncast_bias_weight(...)
This is nessacary for safe async weight offloading.
Currently this peeks ahead to sync the next stream in the queue of streams with the compute stream. This doesnt allow a lot of parallelization, as then end result is you can only get one weight load ahead regardless of how many streams you have. Rotate the loop logic here to synchronize the end of the queue before returning the next stream. This allows weights to be loaded ahead of the compute streams position.
rattus128
force-pushed
the
async-offload-streams
branch
from
a50c8c2 to
6ceba8c
Compare
Done. Thanks. |
|
I think it's good to go - I'll double check with comfy before merging |
1 task
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
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.
Core change notes (2nd commit):
This sync is nessacary as pytorch will queue cuda async frees on the
same stream as created to tensor. In the case of async offload, this
will be on the offload stream.
Weights and biases can go out of scope in python which then
triggers the pytorch garbage collector to queue the free operation on
the offload stream possible before the compute stream has used the
weight. This causes a use after free on weight data leading to total
corruption of some workflows.
So sync the offload stream with the compute stream after the weight
has been used so the free has to wait for the weight to be used.
The cast_bias_weight is extended in a backwards compatible way with
the new behaviour opt-in on a defaulted parameter. This handles
custom node packs calling cast_bias_weight and defeatures
async-offload for them (as they do not handle the race).
The pattern is now:
cast_bias_weight(... , offloadable=True)
thing(weight, bias, ...)
uncast_bias_weight(...)
Example test case:
Linux, RTX3060, 96GB RAM
Wan 2.2 I2V FP8, 192x192x9f, 5+5 steps
python main.py --novram --async-offload
This repeatedly blacks screens and corrupts my outputs:
For control sake, here is the same without --async-offload:
And with this fix:
The function signature change to cast_bias_weight may cause performance regression for users of --async-offload with custom node packs that call this function as a helper. I dont see a way to both keep performance and solve the race without updating custom nodes packs one by one
Amongst my ever growing collection of custom-nodes there are some in @kijai publications (FYI):