On Thu, Apr 11, 2019 at 12:07:12PM +0900, Michael Paquier wrote:
> What do you think about the set attached? I prefer that we use
> psql(on_error_die => 1) so as in the case of a crash or an error the
> test dies hard and fast. In consequence, I have granted
> pg_read_all_settings to the replication user to ensure that no errors
> happen, still the syscache lookup is done. We could have negative
> tests but I don't think that's worth the addition as we already have
> tests for pg_read_all_settings in src/test/regress/. If I remove the
> patched patch from walsender.c, the test fails immediately.
I have done another round of review of this patch and committed it
down to v10. When testing on Windows, I have noticed that these new
regression tests suffer the same problem with SSPI authentication as
the recent tweaks I have done for pg_rewind. So I have backpatched
the extension for PostgresNode::init done in d9f543e9 to be able to
add extra options for the authentication configuration so as the new
tests are able to work.
Thanks Alexander for the report!
--
Michael