[GitHub] [accumulo] DomGarguilo commented on pull request #1888: Fixes #1791 - Flaky test: SuspendedTabletsIT

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

DomGarguilo commented on pull request #1888:
URL: https://github.com/apache/accumulo/pull/1888#issuecomment-773538393


   From your results, this test is still indeed flaky despite my changes. A few of these errors I have never received while running this test on my machine. I will leave my analysis of the errors below:  
   
   **The failed assert on line 253:** 
   The test keeps track of when it can first scan the metadata table after it kills the servers. A comment at this point claims that a tablet can not be suspended before the metadata table can be scanned. It also keeps track of when all suspended tablets are re-hosted. The test asserts that the suspend time is less than the time between those two points. It seems this assert checks to make sure the suspend time is roughly the duration we think it will be (between first scan on metadata table and re-host). Either this is an incorrect assumption, the suspend time is lasting longer than it should, or the tablets are re-hosted before they should be. 
   
   **The failed assert on line 243:**
   Checks that the suspended tablet locations on a server are the same as the tablet locations on that server after it is restarted. This is checking that the tablets are reassigned to the same server in the same location after the server is restarted. If it fails, tablets were either lost, migrated or the way that the test checks this is incorrect. 
   
   **The failed assert on line 225:** 
   This is the most common assert I have received (maybe because it happens first in the test, blocking the others). This occurs when the tablet locations, before and after servers are suspended, differ. This means that the tablets are migrating when they are not supposed to, or the way the test measures the locations is flawed.
   


----------------------------------------------------------------
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