revert to working database system when wal-e-restore fails

[james beedy]

When restoring a new postgresql server from the PITR backup of another fails the new database system is left broken.

Not sure why the wal-e-restore failed, but the database system is left in a broken state following an attempted restore.

$ juju run-action pg-test/0 --wait wal-e-restore storage-uri=s3://pdl-api-prod-postgresql-pitr target-time='2018-07-18 13:22.00.00000+00' confirm=true
unit-pg-test-0:
  id: 80cbbcd5-a23c-40b0-80a3-e2bdaf02c6f4
  status: completed
  timing:
    completed: 2018-07-19 21:08:00 +0000 UTC
    enqueued: 2018-07-19 21:07:45 +0000 UTC
    started: 2018-07-19 21:07:45 +0000 UTC
  unit: pg-test/0

Also, the action status reads 'completed' even though the action failed.

cat /var/log/juju/unit-pg-test-0.log | http://paste.ubuntu.com/p/QtWvdFVfmK/

[Stuart Bishop (stub)]

The recovery_target_time in recovery.conf was set to an invalid value:

    recovery_target_time='2018-07-18 13:22.00.00000+00'

Possibly a locale issue? In any case, it should be validated before writing recovery.conf and destroying the database.

[Stuart Bishop (stub)]

Per example command line, the date PostgreSQL failed to parse was user input. So I think this bug is about adding validation and fixing documentation.

[Stuart Bishop (stub)]

And reporting a failure if we can detect that, rather than a success.