I checked in the new sFlow proxy.
There are just config settings, both optional, so you typically don't need an
sflow { } section at all, just the udp_recv_channel for 6343. However if you
wanted to run sFlow in on an non-standard port such as and you also wanted
to see any VMs that may be reported-on by their hypervisors, then you would
add this to gmond.conf:
udp_recv_channel {
port = /* for sFlow */
}
sflow {
udp_port = /* non-standard port */
accept_vm_metrics = yes
}
All physical (non-VM) metrics will be accepted by default, so the first thing
you will see if you are running host-sflow agents is the inclusion of metrics
such as swap_in and swap_out.
I added documentation to the gmond/conf.pod file, but I don't know if/how this
gets into the man page(?)
Please let me know if there are any problems.
Neil
On Feb 24, 2011, at 4:17 PM, Neil McKee wrote:
On Feb 24, 2011, at 3:07 PM, Bernard Li wrote:
Hi Neil:
I finally had a chance to test out the patch. Didn't run into any
major issue on my end, so +1 from me.
On Thu, Feb 17, 2011 at 2:22 PM, Neil McKee neil.mc...@inmon.com wrote:
There are three metrics to draw particular attention to:
1. System UUID
I noticed that on my Windows host, UUID is
----. I guess UUID generation on
Windows is not supported yet? I tested with hsflowd 1.12 on Windows
XP.
The Windows UUID appears for me, but that's from a Windows7 OS. Maybe we
would need to look in a different place on XP. (The Linux port falls back on
the UUID of the first local disk if it can't get a UUID for the whole system,
so maybe something similar to that would be acceptable as a work around.)
Please raise this question on host-sflow-discuss.
The Datasource ID and Parent Datasource ID can be treated as opaque
strings that the UI could use to capture and represent the containment
hierarchy.
Perhaps you could explain a bit about the format of Datasource ID?
The specs on sflow.org cover this, but basically it consists of
{IPAddress,dsClass,dsIndex} where the IPAddress can be a v4 or v6 address,
and the dsClass tells you if the dsIndex is referring to an interface, a
physical entity or a logical entity (such as a VM or application).
Conceptually I think of each datasource as being one observation point in
the system. From ganglia's perspective it's probably best to treat it as an
opaque string, and just use it to know, for example, that a particular VM
is running on a particular hypervisor.
I've granted you SVN access, so please feel free to check the code
into trunk. But perhaps Brad might want to review the code quickly
before you do so :-)
OK. I'll wait for Brad to comment.
Can you also modify the manpage for gmond.conf plus add it to the
default configuration? I'm okay with accept_all_physical = yes as
the default.
OK.
BTW, are you interested in implementing UUID for gmond? We've been
talking about using UUID instead of hostname/IP as host identifiers
because those things could change, so I think this would be a great
feature to be added to our code base.
I really don't know my way around gmond, and that sounds like it might be a
far-reaching change.
Neil
Thanks,
Bernard
--
What You Don't Know About Data Connectivity CAN Hurt You
This paper provides an overview of data connectivity, details
its effect on application quality, and explores various alternative
solutions. http://p.sf.net/sfu/progress-d2d
___
Ganglia-developers mailing list
Ganglia-developers@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/ganglia-developers
