[release/6.0] Fix stress issues around multiple threads throwing the same exceptions #57959
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
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?s=60&v=4){kind=small}
…s - The watson codebase manipulates the state of the following fields on Exception in a lock-free manner without locks if there are multiple threads throwing the same exception - _stackTrace - _stackTraceString - _remoteStackTraceString - _watsonBuckets - _ipForWatsonBuckets - The designed behavior is that these apis should "mostly" be correct, but as they are only used for fatal shutdown scenarios, exact correctness is not required for correct program execution - However, there are some race conditions that have been seen recently in testing 1. In some circumstances, the value will be explicitly read from multiple times, where the first read is to check for NULL, and then a second read is to read the actual value and use it in some way. In the presence of a race which sets the value to NULL, the runtime can crash. To fix this, the code is refactored in cases which could lead to crashes with a single read, and carrying around the read value to where it needs to go. 2. Since the C++ memory model generally allows a single read written in C++ to be converted into multiple reads if the compiler can prove that the read does not cross a lock/memory barrier, it is possible for the C++ compiler to inject multiple reads where the logic naturally only has 1. The fix for this is to utlilize the VolatileLoadWithoutBarrier api to specify that a read should happen once in cases where it might cause a problem. Finally, the test45929 was tended to fail in GC stress as it would take a very long time to run under GC stress or on some hardware. Adjust it so that it shuts down after about 2.5 minutes. - Do this instead of disabling running under gcstress as there is evidence that there may have been bugs seen during runs under gcstress. Fixes #46803
|
I couldn't figure out the best area label to add to this PR. If you have write-permissions please help me learn by adding exactly one area label. |
jeffschwMSFT
approved these changes
Aug 23, 2021
hoyosjs
approved these changes
Aug 23, 2021
|
@jeffschwMSFT I don't have permissions to merge in this branch. Do we need to set a label so that this gets looked at by a special merge person? |
ghost
locked as resolved and limited conversation to collaborators
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
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.
Backport of #57684 to release/6.0
/cc @davidwrighton
Customer Impact
In very rare situations throwing an exception could cause the runtime to AV accessing NULL. This will FailFast the runtime. However, the issue was found with internally reported failures under GCStress. That failure mode was somewhat more likely, but can only exist on a chk build of the runtime.
Testing
Manual testing of scenario known to cause problems in rare circumstances. It did not repro, and there appears to be no impact to any other CI test. However, as the failure is rather difficult to pin down, its entirely plausible that the issue isn't fixed completely.
Risk
Low.