On Fri, 26 Sep 2025, Laurenz Albe wrote:
On Fri, 2025-09-26 at 12:05 +0000, Dan Mahoney (Gushi) wrote:
In the interest of automation, I've set up a pgpass file for my
pg_basebackup between master and standby. This all works, thusly:
pg_basebackup -d
'postgres://[email protected]:5432/foo?sslmode=verify-ca' -F p
--wal-method=stream -P -R -D /var/db/postgres/data17-test3
However, instead of the password getting baked into the pgsql.auto.conf,
the reference to the passfile gets put in, instead:
# Do not edit this file manually!
# It will be overwritten by the ALTER SYSTEM command.
primary_conninfo = 'user=repuser passfile=''/var/db/postgres/.pgpass''
channel_binding=prefer host=10.1.1.1 port=5432 sslmode=''verify-ca''
sslnegotiation=postgres sslcompression=0 sslcertmode=allow sslsni=1
ssl_min_protocol_version=TLSv1.2 gssencmode=disable krbsrvname=postgres
gssdelegation=0 target_session_attrs=any load_balance_hosts=disable
dbname=foo'
That happens when "pg_basebackup" used a password file to connect to
the PostgreSQL server.
But it seems postgres won't actually read the passfile.
Oh yes, it will, as long as it has permissions 0600, 0400 or 0700 and
belongs to the database server OS user (commonly "postgres").
It must have worked for the "pg_basebackup", so PostgreSQL assumes it
will also work for replication.
I found the problem.
When I did the basebackup, I used a string like:
postgres://[email protected]:5432/foo?sslmode=verify-ca
And my .pgpass file on the replica reflects this:
#hostname:port:database:username:password
10.1.1.1:5432:foo:repuser:xxxx
(I read *somewhere* that you can use a dummy databasename in the .pgpass
file for replication purposes, but the actual DB is ignored.)
What I missed was that for replication, the database name in .pgpass
*must* be 'replication', for pgpass to pay attention to it.
As in, while everything else about the connection string allowed
pgbasebackup to find that line, that same fake DB name was not baked in to
the primary_conninfo allow postgres to find the same user.
-Dan
--
