[
https://issues.apache.org/jira/browse/ACCUMULO-4343?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15330000#comment-15330000
]
Josh Elser commented on ACCUMULO-4343:
--------------------------------------
{quote}
I think this is a result of my hardware being oversubscribed, or just generally
not running on a beefy enough machine.
Instead of comparing medians, maybe we could compare mins, medians, and maxs,
and assert that at least two of the three are ordered? I don't want to play too
many funny games with the statistics here, but I also don't want the test to
fail due to my specific hardware limitations.
{quote}
I'm fairly certain i've also run into problems getting this test to pass
reliably. I'm not 100% sure how best we can fix it though.
> DurabilityIT.testWriteSpeed can fail on slow hardware
> -----------------------------------------------------
>
> Key: ACCUMULO-4343
> URL: https://issues.apache.org/jira/browse/ACCUMULO-4343
> Project: Accumulo
> Issue Type: Bug
> Components: test
> Affects Versions: 1.7.1
> Reporter: Mike Drob
> Assignee: Mike Drob
> Fix For: 1.7.3, 1.8.0
>
>
> When running the ITs, I will sometimes (~30%) get a failure that looks
> something like
> {noformat}
> java.lang.AssertionError: log should be faster than flush
> at org.junit.Assert.fail(Assert.java:88)
> at org.junit.Assert.assertTrue(Assert.java:41)
> at
> org.apache.accumulo.test.functional.DurabilityIT.testWriteSpeed(DurabilityIT.java:103)
> 2016-06-14 04:37:17,774 [functional.DurabilityIT] INFO : Attempt durations:
> [533, 573, 613, 664, 852]
> 2016-06-14 04:37:20,538 [functional.DurabilityIT] INFO : Attempt durations:
> [466, 470, 490, 494, 528]
> 2016-06-14 04:37:23,310 [functional.DurabilityIT] INFO : Attempt durations:
> [433, 464, 500, 512, 513]
> 2016-06-14 04:37:25,447 [functional.DurabilityIT] INFO : Attempt durations:
> [338, 345, 348, 383, 441]
> sync 613 flush 490 log 500 none 348
> {noformat}
> I think this is a result of my hardware being oversubscribed, or just
> generally not running on a beefy enough machine.
> Instead of comparing medians, maybe we could compare mins, medians, and maxs,
> and assert that at least two of the three are ordered? I don't want to play
> too many funny games with the statistics here, but I also don't want the test
> to fail due to my specific hardware limitations.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
