Here is essentially the same patch, but this time with a small bugfix to remove a few lines that were only there for testing.
ganglia_sflow_20110221.patch.gz
Description: GNU Zip compressed data
On Feb 17, 2011, at 2:22 PM, Neil McKee wrote:
> Attached is a patch (against trunk rev 2479) that cleans up the internals of
> the sflow.c file by factoring it out into functions. It separates the XDR
> parsing from the metric-submission, eliminates some rather weak assumptions
> and side-effects, and generally makes it much easier to maintain. In
> particular, it will be easier to add new metrics that may be available soon
> (e.g. from mod-sflow in apache and perhaps memcached).
>
> This patch also adds optional support for the hypervisor and VM metrics that
> hsflowd sends (extracted using the libvirt or Xenstore libs) when it runs on
> a Xen or KVM hypervisor (such as Xenserver, XCP, Proxmox, and RedHat). By
> default these VM fields are left out, but if your gmond.conf has a section
> like this:
>
> sflow {
> null_int = 0
> null_float = 0.0
> accept_all_physical = yes
> accept_all_virtual = yes
> }
>
> then you will see all the VMs showing up. Right now they appear as if they
> were just another host in the cluster, but their metrics are not as
> detailed. For example, you can get VM cpu utilization, but not broken down
> into user/system/idle. You can get VM memory utilization, but not broken
> down into used/cache. The disk and network I/O blocks have more of the usual
> metrics but I gave them separate ids because they are not entirely the same
> as their physical counterparts. For example, you might want to add up
> physical pkts/bytes for the cluster without counting the virtual pkts/bytes.
>
> There are three metrics to draw particular attention to:
>
> 1. System UUID
> 2. Datasource ID
> 3. Parent Datasource ID
>
> A VM may not even have an IP address, or it might change routinely, so I
> used the UUID as the unique identifier. At the moment the hosts hash_table
> is key'd by a field that is a string but is called "IP". If it were called
> "ID" instead then that might be all we need to change. Question: does any
> part of Ganglia rely on the "IP" field actually being an IP Address?
>
> 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.
>
> Just to be clear: none of these new metrics will be included unless you
> explicitly turn them on in the gmond.conf file with the settings above. If
> you turn them off (or run with no sflow{} section at all) then only the
> standard libmetrics fields will be accepted from hsflowd.
>
> Comments?
>
> Neil
> <sflow_20110217.patch>
>
>
>
> On Feb 15, 2011, at 3:14 PM, Bernard Li wrote:
>
>> Hi Neil:
>>
>> On Tue, Feb 15, 2011 at 11:43 AM, Bernard Li <bern...@vanhpc.org> wrote:
>>
>>> I've changed my mind after hearing other people's thoughts on the
>>> matter. The correct fix should be in the frontend code. I'm already
>>> working on it for cpu_report and mem_report. However, it would be
>>> tricky for the load_* metrics. Hopefully we could reach a consensus
>>> on it soon and be able to report load metrics in hsflowd under
>>> Windows.
>>
>> Code committed:
>>
>> https://sourceforge.net/apps/trac/ganglia/changeset/2477
>>
>> Now we just need the load_* metrics from hsflowd Windows and we're in
>> business ;-)
>>
>> Thanks,
>>
>> Bernard
>
------------------------------------------------------------------------------ Index, Search & Analyze Logs and other IT data in Real-Time with Splunk Collect, index and harness all the fast moving IT data generated by your applications, servers and devices whether physical, virtual or in the cloud. Deliver compliance at lower cost and gain new business insights. Free Software Download: http://p.sf.net/sfu/splunk-dev2dev
_______________________________________________ Ganglia-developers mailing list Ganglia-developers@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/ganglia-developers
