feat(kuma-cp) expose config

[jakubdyszkiewicz]

Summary

Starting as draft, because of open questions.

1. Library
   I tried to switch from gopkg.in/yaml.v2 to github.com/ghodss/yaml. It required to change yaml to json annotations, but it turned out that ghodss does not support time.Duration, therefore you cannot put in config stuff like xdsConnectTimeout: 1s. I'd say this is a decrease in UX.

2. Secret values
   For now, we only want to hide the Postgres password, but who knows what will come next. I see 2 ways to implement this.

   1. via annotation like secret:"true" next to envconfig and yaml. Then scan struct recursivly via reflection and replace it via placeholder. Extra tricky part, the value can be of many types.
   2. Just like validation, we can add func HideSecretValues() method on Config, call that for every subconfig and explicitly replace the values.

   I like the second option better. Although probably a bit more code, it is explicit and less magical.

3. Pointers in config
   We have to replace secrets in config so we have to copy a config. It's a problem since subconfigs are pointers. We can "hack" it with marshal -> unmarshal, but I'd say it's better to get rid of pointers. It's kind of uncomfortable to think that any component can change the config at any given time of running app.

[yskopets]

I don't understand. Why don't we want to see this field in YAML ?

[jakubdyszkiewicz]

Otherwise the json/yaml looks wierd. Like this:

{
  "guiServer": {
      "port": 1234,
      "GuiConfig": {
          "ApiUrl": "",
          "Environment": "",
      }
  }
}

everything in config is lowercase instead of this. If we write yaml tags, we will expose this config to the users.

[yskopets]

Up to you.
Personally, I see the value in exposing the entire config even if it comes at a cost.

[yskopets]

I don't understand. Why don't we want to see this field in YAML ?

[yskopets]

I don't understand this.

Here is an example of ghodds/yaml doing marshalling/unmarshalling of time.Duration - https://play.golang.org/p/NdpmTyqitmw

[jakubdyszkiewicz]

Sorry I wasn't clear enough. ghodss supports time.Duration, but you cannot deserialize 2s, you have to write 2000000000 which I'd say is not user friendly at all

[yskopets]

There are workarounds for this, e.g. https://stackoverflow.com/questions/48050945/how-to-unmarshal-json-into-durations

[yskopets]

it turned out that ghodss does not support time.Duration

I'm not sure I get this. Here is a working example - https://play.golang.org/p/NdpmTyqitmw

I like the second option better. Although probably a bit more code, it is explicit and less magical.

Let's go ahead with it. A more conventional name for that func would be Sanitize()

It's a problem since subconfigs are pointers.

Sub-configs are pointers if (and only if) they're optional. It wouldn't be right to break this convention for the sake of /config feature

[yskopets]

There are workarounds for this, e.g. https://stackoverflow.com/questions/48050945/how-to-unmarshal-json-into-durations

[yskopets]

I think, logging API is designed in away to avoid explicit fmt.Sprintf()

[jakubdyszkiewicz]

The problem is that if I the bytes the whole json is escaped which is not readable at all.

[jakubdyszkiewicz]

There are workarounds for this, e.g. https://stackoverflow.com/questions/48050945/how-to-unmarshal-json-into-durations

not sure if custom duration time is better.
I'm gonna merge this as it is, but I'm open for discussion!