Re: [RFC PATCH 0/7] Add script for real-time telemetry monitoring

[Bruce Richardson <bruce.richardson@intel.com>]

On Thu, Dec 11, 2025 at 10:09:10AM +0900, Stephen Hemminger wrote:
> On Wed, 10 Dec 2025 16:55:25 +0000
> Bruce Richardson <bruce.richardson@intel.com> wrote:
> 
> > TL;DR
> > ------
> > 
> > For a  quick demo, apply patces, run e.g. testpmd and then in a separate
> > terminal run:
> > 
> >   ./usertools/dpdk-telemetry-watcher.py -d1T eth.tx
> > 
> > Output, updated once per second, will be traffic rate per port e.g.:
> > 
> > Connected to application: "dpdk-testpmd"
> > Time       /ethdev/stats,0.opackets /ethdev/stats,1.opackets        Total
> > 16:29:12                  5,213,119                5,214,304   10,427,423
> > 
> > 
> > Fuller details
> > --------------
> > 
> > While we have the dpdk-telemetry.py CLI app for interactive querying of
> > telemetry on the commandline, and a telemetry exporter script for
> > sending telemetry to external tools for real-time monitoring, we don't
> > have an app that can print real-time stats for DPDK apps on the
> > terminal. This patchset adds such a script, developed with the help of
> > Github copilot to fill a need that I found in my testing. Submitting it
> > here in the hopes that others find it of use.
> > 
> > The script acts as a wrapper around the existing dpdk-telemetry.py
> > script, and pipes the commands to that script and reads the responses,
> > querying it once per second. It takes a number of flag parameters, such
> > as the ones above:
> >  - "-d" for delta values, i.e. PPS rather than total packets
> >  - "-1" for single-line output, i.e. no scrolling up the screen
> >  - "-T" to display a total column
> > 
> > Other flag parameters can be seen by looking at the help output.
> > 
> > Beyond the flags, the script also takes a number of positional
> > parameters, which refer to specific stats to display. These stats must
> > be numeric values, and should take the form of the telemetry command to
> > send, followed by a "." and the stat within the result which is to be
> > tracked. As above, a stat would be e.g. "/ethdev/stats,0.opackets",
> > where we send "/ethdev/stats,0" to telemetry and extract the "opackets"
> > part of the result.
> > 
> > However, specifying individual stats can be awkward, so some shortcuts
> > are provided too for the common case of monitoring ethernet ports. Any
> > positional arg starting with "eth" will be replaced by the set of
> > equivalent values for each port, e.g. "eth.imissed" will track the
> > imissed value on all ports in use in the app. The ipackets and opackets
> > values, as common metrics, are also available as shortened values as
> > just "rx" and "tx", so in the example above, "eth.tx" means to track the
> > opackets stat for every ethdev port.
> > 
> > Finally, the script also has reconnection support so you can leave it
> > running while you start and stop your application in another terminal.
> > The watcher will try and reconnect to a running instance every second.
> > 
> 
> 
> There was a previous submission of a nice telemetry TUI which seems to have
> been abandoned, is there overlap with this?
> 

Possibly. I didn't remember that being submitted in the past. This was just
something I came up with a while back for my own use, and find it really
handy for quickly getting some perf numbers. Therefore I thought others
might find it of use too.

/Bruce