networkservicemesh / deployments-k8s

Apache License 2.0
42 stars 34 forks source link

Bug with re-deployment of nse-composition with kernel-interfaces #11927

Open isZams opened 6 months ago

isZams commented 6 months ago

Expected Behavior

I would like to change the endpoint composition by adding a new pod in the middle of the chain, by simply modifying the network service definition file.

Actual Behavior

Actually if i add a new pod, there are some problems with the kernel interfaces. Assuming that initially i have this configuration: nsc --> passthrough-4 --> gateway. Then i want to move to this configuration: nsc--> test --> passthrough-4 --> gateway.

The problem with the interfaces is between test and passthrough-4 and then also within passthrough-4. Do you have any idea about the possible reasons? I thought to have a similar behaviour when a client connects to a different service and therefore a new interface is established, as in the example of change-nse-dynamically.

Here there is the zip folder containing all the files that i'm using for this use case. kernel-composition.zip

isZams commented 6 months ago

After several times, the new connections are established.

denis-tingaikin commented 6 months ago

If it's reproducing rarely, it's still a problem. So, does it reproducing?

isZams commented 6 months ago

I checked after 30 minutes and then the interfaces were up. I'm trying to check again if there are problems with the interfaces. Basically the interfaces are created and destroyed after few seconds, until the reach a stable status. From what i remember this behaviour is related to the liveness checks done by the interfaces

denis-tingaikin commented 6 months ago

We want to reach a great quality, so feel free to report any suspected things and many thanks for testing !

isZams commented 6 months ago

Ok, as soon as i get some new information i will inform you. Anyway there are other stuff where i need your help, can you check the other issues?

denis-tingaikin commented 6 months ago

We saw other issues; currently, it requires prioritization from Ed and folks. You may raise these questions on our community calls(next tuesday). See the schedule and links here: https://networkservicemesh.io/community.

isZams commented 6 months ago

Ok. thanks for your reply and your time

isZams commented 6 months ago

After 1 hour i still have the previous problem. I found an error in the configuration file of "test" where i left the old value for the source selector within the cmd-nsc image and i thought that would be the solution but i was wrong.