[
https://issues.apache.org/jira/browse/MESOS-1138?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13961158#comment-13961158
]
Jörn Franke edited comment on MESOS-1138 at 4/5/14 5:02 PM:
------------------------------------------------------------
This sounds very interesting. What about providing some simple rule engine on
the slaves (similarly to the one of udev) to decide what to collect and how to
react (write file?, write HDFS?, write ZooKeeper?). You could also tell in
these rules to execute certain applications (such as perf) and what to do with
the output. Slaves could synchronize these rules with the master(s).
Given this foundation, we can let the user define more complex monitoring
software (e.g. similar to the Netflix Conformity Monkey)
What do you think?
was (Author: jornfranke):
This sounds very interesting. What about providing some simple rule engine on
the slaves (similarly to the one of udev) to decide what to collect and how to
react (write file?, write HDFS?, write ZooKeeper?). You could also tell in
these rules to execute certain applications (such as perf) and what to do with
the output. Slaves could synchronize these rules with the master(s).
What do you think?
> Collect more statistics per container
> -------------------------------------
>
> Key: MESOS-1138
> URL: https://issues.apache.org/jira/browse/MESOS-1138
> Project: Mesos
> Issue Type: Improvement
> Components: containerization, slave
> Reporter: Benjamin Hindman
> Priority: Minor
>
> There are a lot more statistics that we could and should be collecting on
> running containers. For example, some of the statistics available from 'perf'
> would be great additions.
> It's unlikely that we'll be able to figure out an exhaustive list of things
> to collect (especially as new things become available in the future), and
> it's also possible that some operators won't want us to collect all the
> things all the time. To this end, it probably makes sense to make what we
> collect configurable (possibly even dynamically), and even provide some basic
> abstractions that can be shared across containerizers for collecting certain
> statistics.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
