[
https://issues.apache.org/jira/browse/MESOS-8476?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Benjamin Mahler updated MESOS-8476:
-----------------------------------
Description:
Currently, the agent will retrieve the container status upon on each task
status update in order to augment the status update with the container status
information (e.g. ip address). This has made the status update processing
asynchronous when it comes to the side effects to the agent data structures.
Consequently, several bugs have occurred: MESOS-5380, MESOS-7865, MESOS-8459.
It's odd that the container status, which seems to define the properties of the
executor's container, needs to be retrieved in the status update path. Rather,
the agent could just store this once when the executor is launched and remember
it.
Currently, the containerizer interface exposes the container status only as a
separate call. However, to simplify the fix here, the containerizer could
expose it directly in the {{launch()}} Future.
> Store executor container status in the agent after it launches.
> ---------------------------------------------------------------
>
> Key: MESOS-8476
> URL: https://issues.apache.org/jira/browse/MESOS-8476
> Project: Mesos
> Issue Type: Improvement
> Components: agent, containerization
> Environment: Currently, the agent will retrieve the container status
> upon on each task status update in order to augment the status update with
> the container status information (e.g. ip address). This has made the status
> update processing asynchronous when it comes to the side effects to the agent
> data structures. Consequently, several bugs have occurred: MESOS-5380,
> MESOS-7865, MESOS-8459.
> It's odd that the container status, which seems to define the properties of
> the executor's container, needs to be retrieved in the status update path.
> Rather, the agent could just store this once when the executor is launched
> and remember it.
> Currently, the containerizer interface exposes the container status only as a
> separate call. However, to simplify the fix here, the containerizer could
> expose it directly in the {{launch()}} Future.
> Reporter: Benjamin Mahler
> Priority: Major
>
> Currently, the agent will retrieve the container status upon on each task
> status update in order to augment the status update with the container status
> information (e.g. ip address). This has made the status update processing
> asynchronous when it comes to the side effects to the agent data structures.
> Consequently, several bugs have occurred: MESOS-5380, MESOS-7865, MESOS-8459.
> It's odd that the container status, which seems to define the properties of
> the executor's container, needs to be retrieved in the status update path.
> Rather, the agent could just store this once when the executor is launched
> and remember it.
> Currently, the containerizer interface exposes the container status only as a
> separate call. However, to simplify the fix here, the containerizer could
> expose it directly in the {{launch()}} Future.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
