[GitHub] [trafficserver] bdgranger commented on pull request #9303: Issue 9274: read-while-writer readers can't find fragment inserted

[GitBox <gi...@apache.org>]

bdgranger commented on PR #9303:
URL: https://github.com/apache/trafficserver/pull/9303#issuecomment-1387455613

   @masaori335 the WARNINGs happened even before this code change, and are actually to be expected based on the  regression test implementation -- the "insert-delete" test intentionally inserts a Dir entry into every bucket to measure insert and probe rates, but never frees any of the entries. So when the next loop after the "probe rate" message is printed attempts to insert additional entries, it will not be able to find any and it needs to purge due to the directory overflow.  Does the test need to be modified to free some or all of the entries at the end of the insert-delete test, or is it okay to leave those warnings and see that the system deals with the condition appropriately? Either way, that is probably separate from this PR.
   
   However, the hang does appear to be from the code change.  It appears that during the corrupt_bucket test loop, a bucket is selected that was already intentionally corrupted from one of the previous iterations (seems strange that with only 10 loops we would end up back at the same bucket with a rand_CacheKey call), and then the new code that attempts to walk to the end of the list to do the insert gets stuck in an infinite loop.  Need to find the best way to add some protection against that without destroying performance.
   
   Is there any way to stop the build that is still spinning since 3 days ago? I don't appear to have the Jenkins permissions to do that.


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: github-unsubscribe@trafficserver.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org