rpk does not look at redpanda.admin for admin addresses

[jcsp]

When running on a non-default port or address and passing the redpanda config to rpk with --config, one gets an error like:

* unable to fetch metrics from the admin API: Get "http://127.0.0.1:9644/metrics": dial tcp 127.0.0.1:9644: connect: connection refused

It should take the admin endpoint info from the config file.

(This is unrelated to #2751 -- for this issue I had a config file that was in the format rpk understood)

Non default ports will probably be somewhat rare in the wild, but a non-default address could be quite common any time someone wants to make the admin api accessible remotely.

[r-vasquez]

Hello @jcsp can you share your config file please?

I just tested with:

config_file: /etc/redpanda/redpanda.yaml
node_uuid: RkTQpb9sY1QT4ZYAv8V2Jo7XgXQqB2m9mS9rjVTsXniCYvvqw
pandaproxy: {}
redpanda:
  admin:
  - address: 0.0.0.0
    port: 9644
  data_directory: /var/lib/redpanda/data
  developer_mode: true
  kafka_api:
  - address: 0.0.0.0
    name: default
    port: 9092
  node_id: 1
  rpc_server:
    address: 0.0.0.0
    port: 33145
  seed_servers: []
rpk:
  admin_api:
    addresses:
      - 127.44.44.12:1232

and got

* unable to fetch metrics from the admin API: Get "http://127.44.44.12:1232/metrics": dial tcp 127.44.44.12:1232: connect: connection refused

rpk debug bundle assumes 127.0.0.1:9644 when rpk.address.addresses is not set.

[r-vasquez]

Perhaps similar/related with: #4000

[twmb]

Updated the title to reflect this issue's relation to #4000. We can revisit this after a proposed redpanda.yaml redux (#4535)

[jcsp]

My config would not have had a hand-populated rpk.admin_api section (is that documented anywhere?)

[r-vasquez]

I saw that here: https://docs.redpanda.com/docs/reference/node-configuration-sample/ not sure if other docs document this part.

rpk:
...
 # The Admin API configuration
  admin_api:
    # A list of the nodes' admin API addresses that rpk will use.
    addresses:
    - 192.168.72.34:9644
    - 192.168.72.35:9644

[rkruze]

This is a higher priority issue since the admin endpoint is now needed to set cluster configs, which is something I and users of the platform have been hung up on.