[et-mgmt-tools] [PATCH/RFC]: virt-manager: display disk and network I/O statistics

[Cole Robinson]

Guido Günther wrote:
> Hi,
> attached is a more complete version of the net/disk I/O statistics
> patch. Screenshots are here:
>

Awesome! Thanks for this, this has been just waiting to be implemented
for a while :)
 
>  http://honk.sigxcpu.org/misc/vm-overview.png
>  http://honk.sigxcpu.org/misc/vm-detail.png
> 
> There's certainly lots of room for improvement.
> 
>  * all units are KBytes/s, that should be changeable

I would say let that come later. We could either add some way
to make the graphs scalable on demand (if traffic exceeds X KB/sec,
the scale suddenly grows).

Other way would be to add an option to 'Preferences', which really
needs to be broken out into sections. There's a whole bunch of prefs
we could add for stats, like disabling polling outright for the
different data types (since a connection with a ton of VMs will most
likely have scaling issues).

>  * the GtkSparkline graphs display input only. I'd be nice to have in
>    and out in one graph in different colors

Fixing up these graphs to be prettier and more useful is high on
my list. I'm unfortunately not too savvy on custom widgets so there
will be a bit of overhead. gnome-system-monitor has some beautiful
graph widgets, though they are written in C++.

>  * the overview looks a bit boring, should we add sparklines there too,
>    like for cpu usage?

Yes, I would say duplicating the spark lines like you have in the
details section would be sufficient. Maybe just change the heading
in the manager to be 'Disk Activity' and 'Network Activity', and
if the user wants specific data rate numbers they can look at the
individual vm details.

A minor side comment: One of the things we've discussed offline
about adding to virt-manager would be a dedicated performance view
of sorts. It would be a separate window that would allow you to do
all sorts of fun things with toggling and comparing vm stats on the
same graph. We could probably poll for the presence of virt-top
and/or virt-mem to try to scrape them for even more output. So
at some point, we want to have a place for really thorough stats
reporting, which takes the pressure off the current details +
manager view from being really robust.

> 
> I've not cleaned up the disk_usage vs disk_io naming mess yet, since I'd
> wanted to get some feedback first.
> Cheers,
>  -- Guido
> 
> 

> # HG changeset patch
> # User "Guido Günther <agx at sigxcpu.org>"
> # Date 1223040670 -7200
> # Node ID de9d6daefd5ae9b6122ec316816787122d56fc1d
> # Parent  0b86cdc80b2cbeb73fddccf3ea9bcbe72ef94375
> 
> diff -r 0b86cdc80b2c -r de9d6daefd5a src/virt-manager.schemas.in
> --- a/src/virt-manager.schemas.in	Thu Oct 02 19:43:42 2008 +0200
> +++ b/src/virt-manager.schemas.in	Fri Oct 03 15:31:10 2008 +0200
> @@ -66,28 +66,28 @@
>      </schema>
>  
>      <schema>
> -      <key>/schemas/apps/::PACKAGE::/vmlist-fields/disk_usage</key>
> +      <key>/schemas/apps/::PACKAGE::/vmlist-fields/disk_io</key>
>        <applyto>/apps/::PACKAGE::/vmlist-fields/disk_usage</applyto>
>        <owner>::PACKAGE::</owner>
>        <type>bool</type>
>        <default>0</default>
>  
>        <locale name="C">
> -        <short>Show disk usage in summary</short>
> -        <long>Show the disk usage field in the domain list summary view</long>
> +        <short>Show disk I/O in summary</short>
> +        <long>Show the disk I/O field in the domain list summary view</long>
>        </locale>
>      </schema>
>  
>      <schema>
> -      <key>/schemas/apps/::PACKAGE::/vmlist-fields/network_traffic</key>
> -      <applyto>/apps/::PACKAGE::/vmlist-fields/network_traffic</applyto>
> +      <key>/schemas/apps/::PACKAGE::/vmlist-fields/network_io</key>
> +      <applyto>/apps/::PACKAGE::/vmlist-fields/network_io</applyto>
>        <owner>::PACKAGE::</owner>
>        <type>bool</type>
>        <default>0</default>
>  
>        <locale name="C">
> -        <short>Show network traffic in summary</short>
> -        <long>Show the network traffic field in the domain list summary view</long>
> +        <short>Show network I/O in summary</short>
> +        <long>Show the network I/O field in the domain list summary view</long>
>        </locale>
>      </schema>
>  

I haven't tested this, but I don't think we should rename the
gconf field names. There isn't really any benefit, and instead
it will just leave dead fields kicking around.

Changing the description as needed shouldn't be a problem
though.

The rest of the code looks fine. With the above change and
the sparklines in the manager window I'd be glad to take it
in.

Thanks,
Cole