[jira] [Comment Edited] (SENTRY-1471) TestHDFSIntegrationBase.java implements HDFS ACL checking and query verification incorrectly

["Vadim Spector (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/SENTRY-1471?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15494222#comment-15494222 ] 

Vadim Spector edited comment on SENTRY-1471 at 9/15/16 6:56 PM:
----------------------------------------------------------------

Submitting the patch that fixes validation logic. 

a) testAuthzObjOnPartitionMultipleTables() is definitely broken - line 603, permissions not propagated to /tmp/external. Test marked as @Ignore'd
b) testNonHDFSLocations() - in two places calls to verifyOnAllSubDirs() replace "FsAction.ALL" with "null" - - both mentioning SENTRY-546 as justification for code change. They start failing, and reverting to "FsAction.ALL" fixes the test.
c) after the above changes, all tests pass

I suggest proceeding with this JIRA and opening another one, just to fix testAuthzObjOnPartitionMultipleTables() test.


was (Author: vspector@gmail.com):
Submitting the patch that fixes validation logic. 

a) testAuthzObjOnPartitionMultipleTables() is definitely broken - line 603, permissions not propagated to /tmp/external. Test marked as @Ignore'd
b) testNonHDFSLocations() - in two places calls to verifyOnAllSubDirs() replace "FsAction.ALL" with "null" - - both mentioning SENTRY-546 as justification for code change. Well, they actually start failing, and switching back to "FsAction.ALL" fixes the test.
c) after the above changes, all tests pass

I suggest proceeding with this JIRA and opening another one, just to fix testAuthzObjOnPartitionMultipleTables() test.

> TestHDFSIntegrationBase.java implements HDFS ACL checking and query verification incorrectly
> --------------------------------------------------------------------------------------------
>
>                 Key: SENTRY-1471
>                 URL: https://issues.apache.org/jira/browse/SENTRY-1471
>             Project: Sentry
>          Issue Type: Bug
>          Components: Sentry
>            Reporter: Vadim Spector
>            Assignee: Anne Yu
>         Attachments: SENTRY-1471.001.patch
>
>
> verifyOnAllSubDirsHelper() in TestHDFSIntegrationBase.java has a bug. When the number of failed retries exceeds max value (the "else" portion of "if (retry > 0)" statement, it erroneously assigns hasSucceeded = true. It means that verifyOnAllSubDirsHelper() never returns false.
> Once the problem was fixed in a local development branch, several tests in TestHDFSIntegrationAdvanced.java started failing. TestHDFSIntegrationEnd2End tests are still ok.
> Also, it is unfortunate that TestHDFSIntegrationBase() returns boolean instead of throwing the original AssertionError which contains information about expected vs. found ACL permissions - this information is invaluable for debugging. The fix is easy - the to-be-fixed "else" portion is inside the "catch (Throwable th)" clause, so it should just re-throw caught exception. This makes "hasSucceeded" variable unnecessary - verifyOnAllSubDirsHelper() should have "void" return type.
> The same problem exists in verifyQuery(), also  in TestHDFSIntegrationBase.java. The "else" clause of "if (retry > 0)" also assigns "isSucceeded = true;" leading to the same problem - it never fails. Suggested solution would be the same - just re-throw the exception. It will not only fix the problem, but also make debugging easier. Current code "Assert.assertTrue(isSucceeded);" throws away valuable information about the expected vs detected number of database rows.
> Note: the above problems were introduced by SENTRY-1454. Suggested fixes match the original code prior to SENTRY-1454.



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