Terminate services too? about kube-monkey

[asobti]

That sounds fantastic. I would love to see that

[onelapahead]

Awesome, thanks!

[Aergonus]

I think destroying it is unnecessary for services. If you were to rename it, it wouldn't be able to resolve.

[onelapahead]

I like the idea as that would be really simple, but I don't believe you can rename a Service since there isn't a way to update the hostname/URL simultaneously:

$ oc patch svc infinispan -p '{ "metadata": { "name": "infinispanx" } }'
Error from server (BadRequest): the name of the object (infinispanx) does not match the name on the URL (infinispan)
$ kubectl patch svc infinispan -p '{ "metadata": { "name": "infinispanx" } }'
Error from server (BadRequest): the name of the object (infinispanx) does not match the name on the URL (infinispan)

Also even if that was allowed, applications might not respect DNS TTL's, and so they could have the ClusterIP of the Service cached and still be able connect.

I was thinking of two ways of "terminating" the Service:

1. Deleting the Service entirely, but saving a copy of its metadata and spec (and its Endpoint's if it doesn't have a selector i.e. for external IP Service's). Then after a random amount of time restoring the Service.
2. Blackholing the Service through a NetworkPolicy for a random period of time. This would be simpler but it might not work for two reasons I can think of:
   i) NetworkPolicy's have to be enabled on OpenShift by an admin, so a dev can't just simply use kube-monkey right away for Service's
   ii) If NetworkPolicy's were already being used, there might be a rule allowing the app to connect to the Service which would trump the deny all policy

So I'm thinking about going with 1. With either solution, we can reuse kube-monkey/mtbf for the time between failure, and then I was think of kube-monkey/mtbr for "mean time before restoration". I believe those, and the kube-monkey/enabled label will be the only ones needed for Service's.

Thoughts?

[worldtiki]

Closing as stale

Sorry @hfuss. If you still have interest in this please leave a comment.