Breaking change with conninfo_hash with 1.4.4

[ankane]

Hi, it looks there was recently a breaking change with the results of the conninfo_hash method (likely related to #485).

require "pg"

p PG.connect("postgres://localhost/postgres").conninfo_hash.slice(:host, :port, :hostaddr)

1.4.4

{:host=>"localhost,localhost", :port=>"5432,5432", :hostaddr=>"::1,127.0.0.1"}

1.4.3

{:host=>"localhost", :port=>"5432", :hostaddr=>"::1"}

Ref: open-telemetry/opentelemetry-ruby-contrib#142 and ankane/pgsync#169

[cbandy]

I believe the OpenTelemetry code you linked is using the wrong methods to interrogate the active connection. The conninfo and conninfo_hash methods return information about the connection string (possible connections), while the host, hostaddr, port, etc methods return the [one] current/active connection details.

[larskanis]

I agree with @cbandy - the right fix in both cases is to use PG::Connection#host and port instead of dedup the conninfo_hash.

It might be unexpected, to get a different conninfo_hash than arguments given to PG.connect, but we always have been adding some particular items to it by the pg.gem. New is that we resolve the host to multiple IP addresses as described in #485.

[ankane]

Thanks for the responses - will consider going that direction.