[issues/3340] fix for leak #3410
Conversation
finalize head and roll it over in stead of resetting head, this means the current head will always remain a value-null in stead of being reassigned. I applied the fix for the TimeAndSizeBoundReplayBuffer
|
@OlegDokuka - thanks for the review in the other PR, I guess I just completely changed the implementation right as you were reviewing. Anyway this PR only fix the leak. This should be simpler to review and accept :) |
|
this is obviously related to #3340, and fixes the leak, for all cases, including TimeAndSizeBoundReplayBuffer |
|
On a side-note; I have run the stress test that I also added in relation to the PR #3364, and this doesn't fix the concurrency bug, and will still throw null pointers during |
|
@OlegDokuka or others is there anything more I can add here to get this PR through. I haven't had much time for a long time now but it would be nice to push this and possibly the other related one (for the single case) It seems there needs be a license header in the files added, is this simply copy paste work? What about the other failing check I cannot see logs anymore |
|
Closed due to this comment. Let's continue further discussion in the main issue. Thanks. |
This is a simple fix for the leak of memory. It is slightly unsatisfactory as the memory required for the extra node (especially in the
#replay(1)case) is still very high. With the extraNode, taking up 24 unnecessary bytes (27% of the#replay(1)-buffer's memory)I have also added some extra tests, that I initially added for my other attempts at fixing this issue.
I have also fixed the same issue for the
SizeAndTimeBoundReplayBuffer.