I'm trying out DBIx::HA with DBD::Pg. DBIx::HA is intended to help with
high availability, load-balancing and failover.
I noticed a bug when using it with DBD::Pg 1.49. Looking into it
further, I think it could be considered a bug in DBD::Pg.
DBIx::HA builds a look-up table of several connection strings before any
connections are made.
Later, when it needs to use a new database connection, it constructs a
the DSN like this:
my $dsn = 'dbi:'.$dbh->{Driver}->{Name}.':'.$dbh->{Name};
This fails with PostgreSQL, because the original lookup hash uses
"dbname=my_db"
but later $dbh->{Name} returned by DBD::Pg contains just "my_db".
I referenced the DBI docs to see what the standard says about this
value. It says:
Holds the "name" of the database. Usually (and recommended to be)
the same as the ""dbi:DriverName:..."" string used to connect to
the database, but with the leading ""dbi:DriverName:"" removed.
Based on that, I think the assumption in DBIx::HA is reasonable.
Further, I think there is no other way to solve the issue reliably in
DBIx::HA in a driver-independent way, without connecting to all the data
sources first, which misses the point of only failing over as needed.
Mark
