Implement ServerSelector in ServerClaimReconciler

[aobort]

Proposed Changes

• added ClaimedServerRef to .status. Since .spec.serverRef is an optional field it's reasonable to store reference to claimed server in the claim's status
• added handling of .spec.serverSelector field

Fixes #68

[afritzler]

Thanks for the PR @aobort! Please find my comments below.

The core question I have here:
When I claim a Server via a Ref do I want to also allow a Selector or do we ignore it? My understanding of a Ref is: if the Server is Available - claim it! No matter what it's labels are. On the other side the ServerSelector allows you to select the next best Server which has the correct set of labels. So the Ref should have precedence over the selector. On the other side if there is no Ref, try to use the selector.

[aobort]

Thanks for the PR @aobort! Please find my comments above.

The core question I have here: When I claim a Server via a Ref do I want to also allow a Selector or do we ignore it? My understanding of a Ref is: if the Server is Available - claim it! No matter what it's labels are. On the other side the ServerSelector allows you to select the next best Server which has the correct set of labels. So the Ref should have precedence over the selector. On the other side if there is no Ref, try to use the selector.

I'd prefer not to make it overcomplicated and stick to two options:

• use reference if set
• use selector if reference is not set

without implementing complex decision making like "if server by reference is not claimable, then let's try to find another one by selector". In other words, if one put explicit reference to server, then operator should claim it or stop the process.

[afritzler]

You are right: we should stick to the following semantic:

• If .spec.serverRef and .spec.serverSelector is both set, we expect that those labels are present on the Server
• If only .spec.serverRef is set we just pick the Server if it is in an Available state
• If only .spec.serverSelector is set, we List all matching Servers and pick the first Available one

[afritzler]

Can you also run make docs to regenerate the API reference documentation?

[aobort]

@afritzler Test (not included into PR) for the case when neither .spec.serverRef nor .spec.serverSelector defined fails due to ginkgo runs tests in parallel:

it is impossible to define which server will be claimed in case ref and selector are empty. Thus I've made a try to just get server from the claim after .spec.serverRef is fulfilled - works fine. However on some runs - I've run tests 5-10 times in a row - controller claims server which was supposed to be claimed in another test case and that test case fails.

In total: some test case might fail occasionally. The solution which works is to:

1. add Ordered option to the test suite. This will grant serial run for all nested containers, so the test cases will run one by one;
2. destroy servers after each test. This will grant absence of side effects on any objects which are not belong to current test case.

[afritzler]

To test the case, that .spec.serverRef and .spec.serverSelector are empty, just create a couple of Server resources and ensure, that the .spec.serverRef has been mutated to a non nil value. Also you could then try to get the .spec.serverRef object and ensure that it's state is set to Reserved.

In general each suite test (*_test.go) should ensure the correct clean up of any Server which has been created or derived in any of the It statements.

[afritzler]

The PR looks good to me now. Just one minor naming issue I have.

[aobort]

The PR looks good to me now. Just one minor naming issue I have.

@afritzler fixed. Pls, review