ShadingEngine : Add missing this_task_arena::isolate #5402
Closed
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 guess I'll just paste in my repro file here. I've been running it like this:
This should produce a usual runtime of about 1 second scene generation ( plus several seconds loading ), but in the run of 40, you should see some take over 30 seconds and get killed by the timeout command. With the fix, I was able to run a couple of hundred times with no hangs. Contents of deleteFacesHang3.gfr: |
johnhaddon
reviewed
Jul 19, 2023
There was a problem hiding this comment.
it was just a missing this_task_arena::isolate which affected OSLObject.
Something smells fishy here. Although there's a case to be made that ShadingEngine should do an isolation itself, because it could be called from anywhere, it shouldn't be necessary for the call from OSLObject. OSLObject::processedObjectComputeCachePolicy is TaskCollaboration, which means the whole thing should be running isolated in its own arena anyway. Perhaps that's not working? Or perhaps this PR doesn't really fix the root problem, but masks it by changing thread timings somehow?
| auto f = [¶ms, &renderState, &shaderGroup, &shadingSystem, &results]( const tbb::blocked_range<size_t> &r ) | ||
| { | ||
| ThreadInfo &threadInfo = params.threadInfoCache.local(); | ||
| tbb::this_task_arena::isolate( [ ¶ms, &renderState, &shaderGroup, &shadingSystem, &results ]{ |
There was a problem hiding this comment.
I think this could be put around just the tbb::parallel_for call, so it has less overall impact on the code layout, and the meat of the code doesn't end up with an extra level of indentation?
|
Closing, since this doesn't fix the original production problem, and as discussed above isn't necessary because of the |
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.
The hang that's been affecting IE turned out to be completely different from what we were originally thinking - it was just a missing this_task_arena::isolate which affected OSLObject.
I haven't thought that deeply about where this should live - it could go outside ShadingEngine, but it probably needs to affect both OSLImage and OSLObject, so this is probably a reasonable place for it?
I also haven't got a unit test. I do have a fairly clean repro ( Murray has confirmed that it hangs on his computer too ), which I'll try to include here, but it is quite sensitive, and requires being run a large number of times with a large number of threads in order to catch anything.
What would probably be more useful anyway if we were going to improve test coverage would be some way of detecting any tbb parallelism inside TaskCollaborations that isn't isolated ( even if it isn't currently causing a hang ). I started wondering whether we could create a honeypot arena that should never be executed, and could then report an error if anything picks up a job from it? Didn't quite figure out what that would look like though.