Hi all,
(Adding Euler in CC.)
I have mentioned that on this thread:
https://www.postgresql.org/message-id/aOM-T6iojyzb7OVB@paquier.xyz
Attempting to execute pg_createsubscriber compiled with a version of
PG_CONTROL_VERSION incompatible with the standby we want to switch to
a logical replica leads to the following error:
$ pg_createsubscriber -D $HOME/data/5433 \
-P "host=/tmp port=5432" -d postgres
pg_createsubscriber: error: control file appears to be corrupt
For example, setup a cluster based on v17 and run the command. This
is confusing, because the control file is not corrupted,
pg_createsubscriber is just missing the fact that the major version it
is compiled with is not able to manipulate the control file of the
target data folder. We enforce such checks for other tools, like
pg_checksums or pg_rewind, and I don't see why pg_createsubscriber
should be an exception?
One simple thing that could be done is to grab the first line of the
standby's PG_VERSION, at least, so as we have a check that does not
depend on the contents of the control file, and it would unlikely be
corrupted. Having a check based on PG_CONTROL_VERSION would be a few
lines less, of course, that we could try in parallel of the CRC check
report if the value is sane-ish, from a past version.
Thoughts?
--
Michael
