[ 
https://issues.apache.org/jira/browse/MESOS-6825?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15767510#comment-15767510
 ] 

Anand Mazumdar edited comment on MESOS-6825 at 12/21/16 4:54 PM:
-----------------------------------------------------------------

> This means that if the host running the tests is slow, a test case might 
> receive a resource offer that it doesn't receive when running on a faster 
> host.

Did you mean that if the "task" in question in the test reaches a terminal 
state? This shouldn't happen otherwise if the "used" resources on the agent did 
not change (task is active)


was (Author: anandmazumdar):
> This means that if the host running the tests is slow, a test case might 
> receive a resource offer that it doesn't receive when running on a faster 
> host.

Did you mean that if the "task" in question in the test finishes? This 
shouldn't happen otherwise if the "used" resources on the agent did not change 
(task is active)

> Increase default allocation_interval for tests
> ----------------------------------------------
>
>                 Key: MESOS-6825
>                 URL: https://issues.apache.org/jira/browse/MESOS-6825
>             Project: Mesos
>          Issue Type: Improvement
>          Components: tests
>            Reporter: Neil Conway
>              Labels: mesosphere
>
> The default {{allocation_interval}} is 1 second. This means that if the host 
> running the tests is slow, a test case might receive a resource offer that it 
> doesn't receive when running on a faster host. We could workaround this by 
> explicitly using {{WillRepeatedly(Return())}}, but that is a bit kludgy and 
> obscures the intent of the test.
> One way to avoid this would be to pause the clock by default in tests 
> (MESOS-4101). That would be a quite involved change, however.
> Instead, we could consider raising the default {{allocation_interval}} to a 
> large value, such as 1 minute or longer. This would significantly reduce the 
> chance of host performance causing test flakiness. Moreover, it would help to 
> highlight tests that rely on real-time batch allocations for correctness: 
> generally tests should avoid doing this (because it causes test slowness). 
> They should instead pause the clock and then advance it by 
> {{allocation_interval}}.



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

Reply via email to