Use service port name as route targetPort in 'oc expose service'

[liggitt]

Fixes #6265

There are 6 possible scenarios for services with ports we will hit in oc expose

1. 1 unnamed service port -> numbered container port
2. 1 unnamed service port -> named container port
3. 1 named service port -> numbered container port
4. 1 named service port -> named container port
5. 2+ named service ports -> numbered container port
6. 2+ named service ports -> named container port

Whenever a service port is named (cases 3,4,5,6), we should use the name as the route targetPort

When a service port is unnamed (which means it is the only port), we could try to set a targetPort that will point to the right containers, or we could omit the targetPort and let the route "round-robin" to the single port. In case 1 (numbered container port), we could set a numbered targetPort, but the route would break if the service was edited to set a different targetPort. In case 2 (named container port), we have no choice but to round-robin (container port names aren't available in the endpoints).

This PR updates oc expose service <svc> to generate a route that has a named targetPort when possible, and no targetPort otherwise.

[liggitt]

@Kargakis PTAL

[liggitt]

[test]

[liggitt]

ugly, I don't see a better way to explicitly tell the route generator we want it to leave out the port

[0xmichalis]

Can we decide on using port vs target-port as the name of the parameter (and subsequently the name of the flag for setting the route port)? I think the intention is to use target-port. Then we will not require portString to not be zero in the generator, instead of introducing another field to explicitly pass info.

[liggitt]

I think I agree, updated

[0xmichalis]

You also need to update ParamsNames to include target-port (and drop port and ports)

[0xmichalis]

You also need to update ParamsNames to include target-port (and drop port and ports)

And then update the unit tests for the route generator. I always forget those and being reminded by Jenkins is disturbing:)

[liggitt]

updated tests and cmd tests

[0xmichalis]

Is this <no value> what is really returned?

[0xmichalis]

A question, LGTM in general

[liggitt]

that's what template rendering returns when you reference something that doesn't exist in the struct

[0xmichalis]

The command '/bin/sh -c yum install -y libmnl libnetfilter_conntrack openvswitch     libnfnetlink iptables iproute bridge-utils procps-ng ethtool socat openssl     binutils xz kmod-libs kmod sysvinit-tools device-mapper-libs dbus     && yum clean all' returned a non-zero code: 1
!!! Error in hack/build-images.sh:62
    'docker build -t $1:latest $2' exited with status 1
Call stack:
    1: hack/build-images.sh:62 image(...)
    2: hack/build-images.sh:81 main(...)
Exiting with status 1
make: *** [release] Error 1

[merge]

@liggitt thanks for this

[openshift-bot]

continuous-integration/openshift-jenkins/merge SUCCESS (https://ci.openshift.redhat.com/jenkins/job/merge_pull_requests_origin/4334/) (Image: devenv-rhel7_2949)

[0xmichalis]

#6312 flake

[merge]

[openshift-bot]

Evaluated for origin merge up to 419cca5

[openshift-bot]

Evaluated for origin test up to 419cca5

[openshift-bot]

continuous-integration/openshift-jenkins/test SUCCESS (https://ci.openshift.redhat.com/jenkins/job/test_pull_requests_origin/7817/)