Don't have TCP established connection from client but still show connection on the map

[thaonguyen-ct]

Let's say, I have two pods A and B are running on K8s. A(source) sends a TCP request to B(dest) and both have a TCP established connection, but somehow pod A is stopped suddenly and don't send FIN package to close this TCP connection, so the TCP connection is still established on pod B and on pod A has gone. But the response of weave-scope-app API of deployment B still has the inbound connection from A and the response of weave-scope-app API of deployment A still have the outbound connection to B even pod A doesn't send TCP request to B anymore.

What you expected to happen?

Don't see the connection between A and B or at least the don't have the outbound connection on deployment A

What happened?

The connection from A to B still exist on both deployment A and deployment B when call weave-scope-app api

How to reproduce it?

1. Deploy two deployment A and B like Nginx for example
2. From pod A telnet to pod B and you can see the TCP established connection on pod A and B by check netstat -nat
3. From pod A install the iptables and drop outbound of FIN package iptables -I OUTPUT --protocol tcp --tcp-flags FIN FIN -j DROP
4. Kill the telnet command from pod A or crtl-c
5. You can see the TCP connection still established on pod B but not on pod A
6. Call API of weave-scope-app to verify the inbound connection of pod B and outbound connection of pod A still there http:///api/topology/kube-controllers/;%3Cdeployment%3E

Versions:

Weave scope: 1.13.1 Kubernetes: 1.13.3 Docker on kubernetes node: 18.6.3

[bboreham]

It sounds like Scope is telling you what Linux is telling it.

The line on the graph goes away, right? Because one of the process endpoints is now missing.

[thaonguyen-ct]

@bboreham No. The line on the graph is still appear