Yes, I can see that the information i gave in the post was not detailed enough. I do not think it is a formatting error, since the yml works fine without the version tag.
I tested some more, and it seems like whatever bug I encountered does not trigger in version 10 and 11, but does trigger in version 4 to 9.
I can only find two errors in the print, and i think they are expected:
2020-02-22 10:42:33.410 UTC [62] postgres@postgres ERROR: database "discourse" already exists
2020-02-22 10:42:33.533 UTC [73] postgres@discourse ERROR: role "discourse" already exists
I looked at the release notes for version 10, and did not find a reference to a fix related to this.
Its good that it is fixed, and with your work around for the smpt settings (will test it on monday, dont want to mess with the production server during the weekend) there is no urgent need from me, but it would be good to know if there is an unknown error that the tests will not catch, and that may return, if it was ‘fixed’ by accident and not by a targeted change. I guess i want to know how stable the version function can be expected to be in the future.
This is a dump of the output from a run with v2.4.0.beta9 as target version.
dump_disco_version_error.txt (127.5 KB)