[ 
https://issues.apache.org/jira/browse/IMPALA-4128?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16739872#comment-16739872
 ] 

Joe McDonnell commented on IMPALA-4128:
---------------------------------------

I took a glance at this as part of unified backend tests. The difference is due 
to this line 
([be/src/common/logging.cc|https://github.com/apache/impala/blob/a954d5443a613e746f04da9766e77a7a4ea8ce72/be/src/common/logging.cc#L73-L74):]
 ):
{noformat}
// Don't double log to stderr on any threshold.
FLAGS_stderrthreshold = google::FATAL + 1;{noformat}
Promise-test uses google::InitGoogleLogging() directly rather than our 
InitGoogleLoggingSafe() which has the line above. Setting FLAGS_stderrthreshold 
determines whether an error gets logging to stderr. This matters because 
promise-test is matching the output when it fails:
{code:java}
IMPALA_ASSERT_DEBUG_DEATH(
promise.Set(150), "Called Set\\(\\.\\.\\) twice on the same Promise");
{code}
Bufferpool tests don't. e.g. from BufferPoolTest.PinWithoutReservation
{code:java}
IMPALA_ASSERT_DEBUG_DEATH(
discard_result(pool.CreatePage(&client, TEST_BUFFER_LEN, &handle)), "");
{code}
 

> Unusual test setup in promise-test prevents logging into logs/be_tests
> ----------------------------------------------------------------------
>
>                 Key: IMPALA-4128
>                 URL: https://issues.apache.org/jira/browse/IMPALA-4128
>             Project: IMPALA
>          Issue Type: Bug
>          Components: Infrastructure
>    Affects Versions: Impala 2.8.0
>            Reporter: Alexander Behm
>            Priority: Major
>
> In IMPALA-4091 many BE tests were fixed to log into the intended directory 
> logs/be_tests. 
> However, a similar fix did not work for promise-test because changing the 
> test setup caused the test to fail inexplicably (the unusual init order is 
> not commented).
> This JIRA tracks fixing promise-test to log into the correct directory.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to