[
https://issues.apache.org/jira/browse/MESOS-1808?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14143663#comment-14143663
]
Chi Zhang commented on MESOS-1808:
----------------------------------
libnl provides interfaces to get RTT information from its diagnosis module
(idiag). This should give us a few advantages over using `ss` directly: 1)
don't have to depend on ss output format. It makes code agnostic to OS
environment / ss versions, etc 2) better scalability since there is no printing
and slow string parsing. prior investigation has found that the number of
active connections in a container could be as high as ~200,000 3) more
extensible. more diagnosis information is available in the idiag module should
we want them: socket mem info, re-transmissions, etc.
step 1:
Modify the routing library to reflect libnl idiag module. I will start simple
by only including what's needed for this ticket but make it generic in case we
need more in the future. I am working on this part.
step 2:
add necessary logic to Network Isolator to report RTT out of 'usage()'. I will
probably use the 'Subcommand' primitive to 'setns' into the container and
return requested data via JSON.
> expose RTT in container stats
> -----------------------------
>
> Key: MESOS-1808
> URL: https://issues.apache.org/jira/browse/MESOS-1808
> Project: Mesos
> Issue Type: Task
> Components: containerization
> Reporter: Dominic Hamon
> Assignee: Chi Zhang
>
> As we expose the bandwidth, so we should expose the RTT as a measure of
> latency each container is experiencing.
> We can use {{ss}} to get the per-socket statistics and filter and aggregate
> accordingly to get a measure of RTT.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
