"Nick Renders" <postgres@xxxxxxxxxx> writes:
> [ pg_upgrade fails with ]
> connection to database failed: fe_sendauth: no password supplied
> could not connect to source postmaster started with the command:
> "/Library/PostgreSQL/11/bin/pg_ctl" -w -l "pg_upgrade_server.log" -D
> "/Volumes/Postgres_Data/PostgreSQL/11/data" -o "-p 49156 -b -c
> listen_addresses='' -c unix_socket_permissions=0700 -c
> unix_socket_directories='/Volumes/Free/Upgrade'" start
> According to the documentation, the connection should be established
> with the data in the .pgpass file. Its contents look like this (the
> password has been changed) :
> localhost:49155:*:postgres:password1234
> localhost:49156:*:postgres:password1234
I think this is explained by this statement in the libpq documentation:
The host name localhost is also searched for when the connection is a
Unix-domain socket connection and the host parameter matches libpq's
default socket directory path.
pg_upgrade will use a Unix-domain socket (unless on Windows), but it
intentionally puts it in a non-default place --- we can see
unix_socket_directories='/Volumes/Free/Upgrade'
in your example. That's meant to ensure that outside clients can't
connect to the postmaster(s) during the upgrade, but it's not interacting
too well with this behavior of libpq.
I don't recall for sure, but I think you could have made this work
by putting the socket path (/Volumes/Free/Upgrade) instead of
"localhost" in the .pgpass file.
regards, tom lane
