[
https://issues.apache.org/jira/browse/AURORA-1658?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Stephan Erb updated AURORA-1658:
--------------------------------
Description:
Aurora is declining Mesos offers in various scenarios:
* when launching a task, the remaining resources of the offer are declined
implicitly
* when compacting multiple offers of a slave into a single one
Whenever we do this, we don't pass a {{Filter}} and [therefore use the default
of 5
seconds|https://github.com/apache/mesos/blob/df29bf0338771c92d1b1d3848181a35429cdcf0f/include/mesos/mesos.proto#L1393-L1400].
This tells Mesos to return the resources to us only after a timeout of 5
seconds, even if there is no other framework that wants them.
This is not a problem if you have large clusters. However for small clusters or
clusters with little available free resources this can lead to Aurora
unnecessarily throttling itself.In addition, this might lead to tasks being
scheduled on nodes which have be been put into maintenance mode.
was:
Aurora is declining Mesos offers in various scenarios:
* when launching a task, the remaining resources of the offer are declined
implicitly
* when compacting multiple offers of a slave into a single one
Whenever we do this, we don't pass a {{Filter}} and [therefore use the default
of 5
seconds|https://github.com/apache/mesos/blob/df29bf0338771c92d1b1d3848181a35429cdcf0f/include/mesos/mesos.proto#L1393-L1400].
This tells Mesos to return the resources to us only after a timeout of 5
seconds, even if there is no other framework that wants them.
This is not a problem if you have large clusters. However for small clusters or
clusters with little available free resources this can lead Aurora
unnecessarily throttling itself.In addition, this might lead to tasks being
scheduled on nodes which have be been put into maintenance mode.
> Aurora is throttling itself via a non-optimal offer filter time
> ---------------------------------------------------------------
>
> Key: AURORA-1658
> URL: https://issues.apache.org/jira/browse/AURORA-1658
> Project: Aurora
> Issue Type: Bug
> Components: Scheduler
> Reporter: Stephan Erb
> Assignee: Stephan Erb
> Labels: perfomance
>
> Aurora is declining Mesos offers in various scenarios:
> * when launching a task, the remaining resources of the offer are declined
> implicitly
> * when compacting multiple offers of a slave into a single one
> Whenever we do this, we don't pass a {{Filter}} and [therefore use the
> default of 5
> seconds|https://github.com/apache/mesos/blob/df29bf0338771c92d1b1d3848181a35429cdcf0f/include/mesos/mesos.proto#L1393-L1400].
> This tells Mesos to return the resources to us only after a timeout of 5
> seconds, even if there is no other framework that wants them.
> This is not a problem if you have large clusters. However for small clusters
> or clusters with little available free resources this can lead to Aurora
> unnecessarily throttling itself.In addition, this might lead to tasks being
> scheduled on nodes which have be been put into maintenance mode.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
