[BUG] epicli apply fails when rerun for cluster on which 'repmgr standby switchover' was performed

[to-bar]

Describe the bug
Epiphany assumes and requires for PostgreSQL cluster that the first inventory host is always primary node and the second standby. If roles are switched, it fails with error:

TASK [postgresql : Register primary with repmgr]
skipping: [ec2-54-208-80-28.compute-1.amazonaws.com]
fatal: [ec2-54-172-24-132.compute-1.amazonaws.com]: FAILED! => {"changed": true, "cmd": "/usr/bin/repmgr primary register -f /etc/postgresql/10/main/repmgr.conf --force --superuser=epi_repmgr_admin", "delta": "0:00:00.047723", "end": "2020-11-12 14:20:29.793961", "msg": "non-zero return code", "rc": 1, "start": "2020-11-12 14:20:29.746238", "stderr": "ERROR: server is in standby mode and cannot be registered as a primary", "stderr_lines": ["ERROR: server is in standby mode and cannot be registered as a primary"], "stdout": "", "stdout_lines": []} 

To Reproduce
Steps to reproduce the behavior:

1. Create cluster with PostgreSQL replication enabled.
2. Switchover database nodes, see how-to doc.
3. Run epicli apply again (the same config as in step 1).

Expected behavior
epicli apply should not fail and allow to have switched roles.

Config files
n/a

OS (please complete the following information):

• Found on Ubuntu

Cloud Environment (please complete the following information):

• All

Additional context
n/a

[mkyc]

Linked with #2074

[to-bar]

This issue has been solved in PostgreSQL upgrade related feature branch but we should re-test after merging code to develop branch.

[przemyslavic]

✔️ Fixed.
Successfully deployed a 3-node cluster, performed a switchover and executed epicli apply again.