From: Michael Paquier
Hm.. I have just tested HEAD, my patch and your patch using my patch
test on pg_ctl.c, but I am always getting pgwin32_is_service set to 0
when running pg_ctl start from a terminal, and set it to 1 when
running pg_ctl service to register the service startup. Could you
precise in which ways you started the Postgres instance and could you
post the patch of pg_ctl you used? I am afraid that I am taking it
incorrectly because I am not able to see any differences.
Also, did you test the patch I posted and were you able to see the
same differences as with your patch? I still think that my short patch
is logically correct but if the tests are not we are in a no-go
position for any fix posted on this thread.
Yes, I tested both your patch and mine. I used the attached pg_ctl.c.
It adds -z option which disables SECURITY_SERVICE_RID.
I registered the service with "pg_ctl register -N pg -D
datadir -w -z -S demand -U myuser -P mypass", then started the service
with "net start pg". The following messages were output in the server
log:
LOG: pgwin32_is_admin = 0
LOG: pgwin32_is_service = 0
LOG: database system was shut down at 2016-11-07 22:04:46 JST
LOG: MultiXact member wraparound protections are now enabled
LOG: database system is ready to accept connections
LOG: autovacuum launcher started
Without -z, the message becomes "pgwin32_is_service = 1". And without
the win32security.c patch, "pgwin32_is_service = 1" is output.
I guess you registered the service without specifying the service
account with -U. Then the service runs as the Local System account,
whence pgwin32_is_service() returns 1.
Regards
Takayuki Tsunakawa
