Add /config handler which exposes the current YAML config of cortex.

[tomwilkie]

Signed-off-by: Tom Wilkie tom@grafana.com

Fixes #1886

Checklist

• Tests updated
• Documentation added
• CHANGELOG.md updated - the order of entries should be [CHANGE], [FEATURE], [ENHANCEMENT], [BUGFIX]

[cyriltovena]

Nice, We need this in Loki too and also add overrides via /config/{tenant_id}/overrides.

LGTM

[Wing924]

I think we need to mask passwords.

[cyriltovena]

I think we need to mask passwords.

Good point, although I think the usage is meant to be internal.

[pracucci]

Thanks Tom for this! I would love to have it as well.

I gave it a try and unfortunately the generated config doesn't work if you try to use it (ie. you get cannot unmarshal !!map into string) because the marshalling of some custom data types is missing or incorrect. I personally would expect to be able to copy-paste the config and get it working.

Few examples of data types with this issue:

flagext.URLValue:

  externalurl:
    url:
      scheme: ""
      opaque: ""
      user: null
      host: ""
      path: ""
      rawpath: ""
      forcequery: false
      rawquery: ""
      fragment: ""

url.URL:

  externalurl:
    url: null

In this regards, an integration test which runs a component (ie. distributor), gets the /config, save the config to file and re-run the same component with the store config could help to spot regressions over the time.

I think we need to mask passwords.

What if we add a config option to enable this API while keeping it disabled by default?

[pracucci]

@tomwilkie I've pushed some commits. Please take a look.

In particular, the omitempty across the config structs causes some troubles with this API. For example, if in the config you set min_ready_duration: 0s, given 0s is the zero value, the min_ready_duration will not be included in the exported config and loading the exported config will load the default value (1m). I think it could be worth to get rid of all omitempty. What's your take?

[tomwilkie]

Thanks @pracucci! Your commits LGTM. Good catch re:omitempty, yes I think these should all be removed - and there are a lot of them.

[pracucci]

re:omitempty, yes I think these should all be removed - and there are a lot of them.

Yes, but we can postpone it to a separate PR.