[jira] [Created] (ACCUMULO-4185) Race condition in ProxyDurabilityIT

["Josh Elser (JIRA)" <ji...@apache.org>]

Josh Elser created ACCUMULO-4185:
------------------------------------

             Summary: Race condition in ProxyDurabilityIT
                 Key: ACCUMULO-4185
                 URL: https://issues.apache.org/jira/browse/ACCUMULO-4185
             Project: Accumulo
          Issue Type: Bug
          Components: test
            Reporter: Josh Elser
            Assignee: Josh Elser
            Priority: Minor
             Fix For: 1.7.2


Noticed a failure in ProxyDurabilityIT on $dayjob test infrastructure.

When using {{Durability.LOG}} on a condition update through the proxy, we did not see the update after restarting the tabletserver (in MAC).

I talked to [~ecn] in chat about this because he had made a change as a part of his work in ACCUMULO-3423 in https://github.com/apache/accumulo/commit/3fdd29f5222f9d1d32ca28b5ecf1d740a8d20f87

Turns out that there is a subtle race condition when using LOG and killing the TabletServer with RawLocalFileSystem. It's possible that even though we wrote an update with LOG durability (it hit the WAL), because we didn't sync it, the update might be lost after a restart. Key word: "might".

A quick fix is to pull back the change that Eric made to use SYNC and always expect the update to be present.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)