[GitHub] [avro] RyanSkraba commented on pull request #1078: AVRO-3033: Fix flaky file descriptor test

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

RyanSkraba commented on pull request #1078:
URL: https://github.com/apache/avro/pull/1078#issuecomment-771590537


   Hello!  When I repeatedly ran the test, there were 5-10 failures over 100K runs in 20 seconds.  That's not really rigorous, because I wasn't running in any sort of isolation and my machine was doing other stuff.  There were never two failures in a row.
   
   It looks like before fixing the bug AVRO-2286, the number of openFilesBeforeOperation was **always** greater than openFilesAfterOperation.  In practice, now they're **usually** the same value, which is promising evidence that there isn't a resource leak there.  
   
   The flakiness of the test comes from the fact that "something" might be fuzzing either of them off by one.  I used to see this in travis and local builds too (but infrequently).  It might be a background, system-process or an action like garbage collection.  I don't think it's instability of the machine, it's just not as deterministic as we'd like.
   
   I'll add this info to the commit message and we can watch for further failures, I guess!  If we find evidence for either resource leaks or we see the flaky test continue, we can revisit it.


----------------------------------------------------------------
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.

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