Fixes network observer traffic metrics bug

[c-kruse]

Previously traffic metrics (skupper_sent_bytes_total) was only updated when skupper_received_bytes_total was also incremented. This is especially visible in asymetric applications when data flows in one direction like iperf.

Fixes #2309

[c-kruse]

@pwright published a test image.

If you want to try this out yourself you need these flags with helm --set image.repository=quay.io/ckruse/skupper/network-observer --set image.tag=gh-2309.

You could run it side-by-side with an existing (broken) version like this:

helm install my-fixed-release \
   oci://quay.io/skupper/helm/network-observer \
   --version 2.1.2 \
   --set image.repository=quay.io/ckruse/skupper/network-observer \
   --set image.tag=gh-2309

Or upgrade the existing one

helm upgrade skupper-network-observer \
  oci://quay.io/skupper/helm/network-observer \
  --set image.repository=quay.io/ckruse/skupper/network-observer --set image.tag=gh-2309

[pwright]

just did a quick test, data is now showing at component level (but doesn't look complete). Process data is still blank
Attached some random screenshots

[c-kruse]

Hi @pwright, that looks to me like a classic example of a confused network observer, and think we've probably fixed just one small slice of traffic data not working here.

From bottom to top:
I see a site-client-10.244.x.x "process" with a ~15s lag between connections an data outbound. This is pretty much exactly what I'd expect when the observer doesn't know who that address (10.244.x.x) is. It sees some traffic from that address, and waits a bit (exactly 15s) then gives up and tosses it under a placeholder.

components/simple-iperf shows only incoming traffic in the server role, and no outgoing bytes as a client. We'd expect that, except the observer did not manage to figure out who that traffic was coming from and made up a placeholder, so it never gets attributed to simple-iperf.

Above this, I kind of lose the thread on what I'm looking at. After the past nearly two years, these pages all still look the same to me.

[pwright]

@c-kruse so not related to original issue, thx