add serializeable start config

[deads2k]

More unit tests need to be added, but you can start to see the bones.

Before merging:

• DNS changes in dc16de5
• CA changes in Add cert validation option to requestheader #1246
• re-enable integration dns test
• standardize method receiver names
• fix weird cert serial number problem
• serialize master/node startup
• fix certArgs location (not logically part of masterArgs or nodeArgs). Not worth the effort. We use those arguments to build the default cert locations. Rather than expand all the command line args, we'll just keep the certArgs.
• godoc

Next list:

• update integration tests to use the intermediate config
• write out bootstrap roles and role bindings
• Divide config into stanzas that represent the things we could spawn in independent processes
• Mint users for router and registry
• Validate config
• make auth config
• update usage for mint commands to list all the flags
• update vagrant to mint certs ahead of time
• resolve relative file references in the master/node config files relative to the files

@liggitt

[liggitt]

not sure what the behavior of user.GetName() is if UserFrom says it didn't exist

[deads2k]

@liggitt don't freak out. I just had to burn the village to save it. /server is dead, long live /startconfig

[liggitt]

don't make this look like a real username... make it "Unknown user" or something

[deads2k]

Alright, I'm happy with the shape of this now. It allows you to separately mint certs. All in one launch delegates down to master and node like you would expect. It can generate the fully specified for configs for you and read them back in.

I need to fix up unit tests and integration tests, but I think I like the final shape of things.

Next steps for future pulls:
[ ] write out bootstrap roles and role bindings

There was more to that list when I started it.....

[liggitt]

List of things to make sure don't get dropped:

• DNS changes in dc16de5
• CA changes in Add cert validation option to requestheader #1246

Suggestions for future pulls:

• Divide config into stanzas that represent the things we could spawn in independent processes

[liggitt]

It's super confusing for kubernetes.NodeConfig.NodeHost to not come from configapi.NodeConfig.NodeHost. I actually prefer NodeID, but can we name the other one BindHost?

[deads2k]

Done. NodeID versus Hostname is going to an upstream back down to us thing. Otherwise it's always confusing in some location or another.

[deads2k]

rebased

[liggitt]

Seems like this should not complain about DNS, and should tell you a master and node config path are required:

$ openshift start --config-only=true
W0316 13:55:18.330605    7974 master_args.go:297] Unable to bind DNS on port 53 (you may need to run as root), using { %!s(int=8053) %!s(bool=false) %!s(bool=false) 0.0.0.0 tcp://0.0.0.0:8053 0.0.0.0 %!s(int=8053)} which will not resolve from all locations
F0316 13:55:18.331725    7974 start_allinone.go:75] open : no such file or directory

[liggitt]

lowercase leading

[deads2k]

rebased

[liggitt]

this won't resolve file references relative to the filepath :-/

[deads2k]

We'll get that working with kubernetes/kubernetes#5541. For now, the generated .kubeconfig files are self contained.

[deads2k]

Using the upstream patch for now.

[liggitt]

all-in-one isn't setting ClusterDNS on node args to the master IP

[liggitt]

Opened deads2k#6 to improve DNS handling:

• Bind defaults to 0.0.0.0:53 when simply writing out config
• Never switch binding from :53 to :8053 at runtime, just print a warning about DNS not initializing
• Integration tests have the ability to customize the DNS bind port for testing (to avoid needing to run as root)
• Correctly wire master IP to node DNS IP in all-in-one
• Correctly serialize dns IP in node config
• Rename DNSAddress to DNSIP in node config
• Print out the name of the running integration test before starting it so you can see which test is hanging
• Fixed login integration test compile issue

@smarterclayton can you eyeball the DNS changes in my pull?

[liggitt]

relative file references in the master/node config files get resolved relative to pwd, not relative to the file... feels like we should be consistent with kubeconfig (relative to config file), but can be done as a follow-up

[liggitt]

LGTM

[liggitt]

[merge]

[openshift-bot]

continuous-integration/openshift-jenkins/merge SUCCESS (https://ci.openshift.redhat.com/jenkins/job/merge_pull_requests_openshift3/1196/) (Image: devenv-fedora_1061)

[openshift-bot]

Evaluated for origin up to 3038033