[
https://issues.apache.org/jira/browse/MESOS-1228?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13976922#comment-13976922
]
Timothy St. Clair edited comment on MESOS-1228 at 4/22/14 3:39 PM:
-------------------------------------------------------------------
If you route a NAT though iptables you can always get stats ;-)
ref: https://github.com/openshift/geard/blob/master/docs/linking.md
If you go a step further, and maintain global state with either zookeeper or
etcd, then publishing and service discovery become a snap, vs. jigging state
inside of a scheduler.
was (Author: tstclair):
If you route a NAT though iptables you can always get stats ;-)
ref: https://github.com/openshift/geard/blob/master/docs/linking.md
If you go a step further, and maintain global state with either zookeeper or
etcd then publishing and service discovery become a snap, vs. jigging state
inside of a scheduler.
> Container level network monitoring
> ----------------------------------
>
> Key: MESOS-1228
> URL: https://issues.apache.org/jira/browse/MESOS-1228
> Project: Mesos
> Issue Type: Story
> Components: containerization, isolation, statistics
> Affects Versions: 0.19.0
> Reporter: Jie Yu
> Fix For: 0.19.0
>
>
> Our goal is to get per-container network statistics. For instance, number of
> packets received/sent (RX_PACKETS, TX_PACKETS), how many bytes received/sent
> (RX_BYTES, TX_BYTES), etc.
> We have a couple of options here for implementing it.
> 1) If we have enough IPs (we are assuming IPv4 here as most people are still
> using IPv4) such that each container can has an individual IP, we can
> leverage the Linux bridge.
> 2) If we don't have enough IPs, we can use port forwarding based on port
> ranges. In other words, each container will be assigned a range of ports from
> the host. The isolator will setup filters to properly redirect packets to the
> corresponding containers based on dst ip/ports.
> 3) We can use NAT if the processes running inside a container does not need
> public IPs.
> Our first step is to go with option (2).
--
This message was sent by Atlassian JIRA
(v6.2#6252)
