[jira] [Commented] (MESOS-9616) `Filters.refuse_seconds` declines resources not in offers.

["Chun-Hung Hsiao (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/MESOS-9616?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16779897#comment-16779897 ] 

Chun-Hung Hsiao commented on MESOS-9616:
----------------------------------------

It seems an unexpected behavior when a reserved resource or created PV gets declined just because a framework is not ready to use the resource yet. This might not be a big problem in the past, but with resource providers capable of adding resources dynamically, frameworks might want to first reserve one resource and wait for another to show up before launching a task to consume both.

> `Filters.refuse_seconds` declines resources not in offers.
> ----------------------------------------------------------
>
>                 Key: MESOS-9616
>                 URL: https://issues.apache.org/jira/browse/MESOS-9616
>             Project: Mesos
>          Issue Type: Bug
>          Components: allocation
>            Reporter: Chun-Hung Hsiao
>            Priority: Major
>              Labels: mesosphere, resource-management
>
> The [documentation|http://mesos.apache.org/documentation/latest/scheduler-http-api/#accept] of {{Filters.refuse_seconds}} says:
> {quote}
> Also, any of the offer’s resources not used in the ACCEPT call (e.g., to launch a task or task group) are considered declined and might be reoffered to other frameworks, meaning that they will not be reoffered to the scheduler for the amount of time defined by the filter.
> {quote}
> Consider an {{ACCEPT}} call with just a {{CREATE}} operation, but no {{LAUNCH}} or {{LAUNCH_GROUP}}. The {{CREATE}} call will generate a persistent volume resource that is *not* in the offer's resources, but it will still not be reoffered to the scheduler for the amount of time defined by the filter.
> Also, the term *used* is vague here. If we have an {{ACCEPT}} call with a {{CREATE}} on a disk followed by a {{DESTROY}} on the created persistent volume, would the disk be considered *used*?



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